An empirical investigation of the effects of intra-EU BITs on FDI flows within the EU

Faculty of Economics and Business

An Empirical Investigation of the Effects of Intra-EU

BITs on FDI Flows within the EU

Master Thesis International Economics and Globalization Dennis van Zanten

Student Number: 10080813 Supervisor: Prof. dr. D.J.M. Veestraeten Second Reader: Prof. dr. F.J.G.M. Klaassen

Date: 15-07-2016 Words: 15194

Abstract

Bilateral investment treaties (BITs) grant foreign investors the procedural rights to enforce the treaty provisions in international arbitration courts. When the treaty provisions are broken by a host-country’s government, compensation can be demanded for the loss in profits. This extra protection is expected to attract more foreign direct investments by host-countries. To protect their investors, the European Union (EU) members signed many of these BITs with countries in Central and Eastern Europe (CEEC). The accession of the CEEC to the EU created intra-EU BITs, which are conflicting with the European Community Law (ECL) and causing discrimination on the grounds of nationality, because not every investor in the EU has access to the extra protection offered by the treaties. Intra-EU BITs seem to offer better protection than ECL when it comes to expropriation of foreign investments within the EU. Some EU member states have refused the official request of the European Commission to terminate the intra-EU BITs. These countries view the treaties as competitive advantages for their investors, who actively engage in arbitration to protect their interests. The empirical results of this paper suggest that the intra-EU BITs do not have a significant influence on FDI inflows within the EU.

1

Statement of Originality

This document is written by student Dennis van Zanten who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is 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.

2

Table of contents:

List of Abbreviations... List of Figures...

1. Introduction... 2. BITs and their impact on FDI...

i.

BITs and FDI...

ii.

Intra-EU... 3. Empirical application...

i.

Literature...

ii.

Methodology...

iii.

OLS Results...

iv.

GMM-system results... 4. Conclusion... 5. References... 3 4 5 8 8 11 17 17 24 31 40 43 45

3

List of Abbreviations:

AR(1) Arellano-Bond test for first-order serial correlation AR(2) Arellano-Bond test for second-order serial correlation BIT Bilateral Investment Treaty

CEEC Central and Eastern European Countries CPI Consumer Price Index

ECJ European Court of Justice

ECL European Community Law

EU European Union

EUC European Commission

FDI Foreign Direct Investment GMM General Methods of Moments

MIGA Multilateral Investment Guarantee Agency MNE Multinational Enterprise

ICRG International Country Risk Guide

ICSID International Centre for the Settlement of Investment Disputes ISDS Investor State Dispute Settlement

IV Instrumental Variable

OECD Organisation for Economic Co-Operation and Development OLS Ordinary Least Squares

4

List of Figures:

Figure 1: The rise of BITs worldwide... Figure 2: Number of intra-EU BITs per EU member... Figure 3: EU countries with the number of active intra-EU BITs... Figure 4: Known ISDS cases to date... Figure 5: Country-sample ... Figure 6: Different dependent variable specifications... Figure 7: Data sources... Figure 8: Expected signs... Figure 9: OLS regression results : No averages... Figure 10: OLS regression results : Two year averages... Figure 11: OLS regression results : Three year averages... Figure 12: OLS regression results : Four year averages... Figure 13: OLS regression results : Three year averages... Figure 14: GMM-system regression results : Three year averages...

5 11 12 15 25 26 30 31 32 33 34 35 38 41

5

1. Introduction

The European Union (EU) gained the exclusive competence on foreign direct investment (FDI) regulations after the Lisbon Treaty went into force in 2009. FDI was one of the last key areas of commercial activity that had not been part of the collective community regime. Its inclusion enables the EU to benefit from increased bargaining power in (future) negotiations on investment agreements (Schicho, 2012). Prior to the Lisbon Treaty, the individual EU members concluded many bilateral investment treaties (BITs) with countries outside the EU in order to protect their FDI abroad (EUC, 2010). From figure 1 below it becomes clear that the EU countries are involved in more than half of all the BITs worldwide. Practical concerns regarding some of these BITs arose when countries from Central and Eastern Europe (CEEC) joined the EU in 2004, 2007 and 2013. With their accession the first intra-EU BITs were created, because many of the new participants had concluded BITs with the existing EU members earlier on (Eilmansberger, 2009). As of to date there are 209 intra-EU BITs in force, which are represented by the blue line in figure 1 below.

Figure 1: The rise of BITs worldwide. Source: UNCTAD, 2016.

BITs include rules and standards on the admission, treatment, expropriation, and the settlement of disputes regarding FDI (Elkins et al., 2006). The most important settlement mechanism is the Investor-State Dispute Settlement (ISDS), which provides investors from the treaty partner-country with the procedural rights to sue the host government in international arbitration courts, if any of

6 the treaty obligations are broken. Here monetary compensation for the loss in profits resulting from the host state breaking its obligations can be demanded directly (Allee & Peinhart, 2010).

The European Commission (EUC) is of the opinion that these intra-EU BITs have become obsolete after the enlargement of the EU, because all investments within the single market should be subject to the same rules and protection provided by European Community Law (ECL) (EUC, 2015). Dimopoulos (2011) acknowledges that many BIT provisions overlap with ECL. However, ECL for instance does not include provisions on the protection of FDI of EU nationals in other EU member-states against political risks and expropriation. Guzman (1997) mentions that political risks follow from the time-inconsistency of government behaviour, whereby government policies to attract FDI are no longer optimal after the FDI has been made. This time-inconsistency can be seen as discouraging FDI, because it creates uncertainty and raises expected costs. BITs are designed to overcome this time-inconsistency by allowing investors to enforce the BIT provisions also after the initial investment has been made. Within the EU the time-inconsistency is still present, since member-states are allowed to change national laws within the margins of the ECL, which is subject to changes itself as well (Dimopoulos, 2011). BITs restrict the ability of EU member-states to alter FDI regulations later on, because they traded a degree of sovereignty for credibility. The removal of the time-inconsistency for FDI is why intra-EU BITs still provide higher standards of protection than the ECL (Wehland, 2009). Some Member-states see the extra protection and ISDS as competitive advantages for their prospective and actual foreign direct investors (Ghouri, 2010).

This competitive advantage might lead to discrimination of investors on the grounds of nationality, which does not comply with the principle of equal treatment within the EU (Ghouri, 2010). In addition, intra-EU BIT members are facing numerous ISDS claims on the ground that regulatory changes to comply with ECL are conflicting with the BIT obligations (UNCTAD, 2015). Furthermore, when investors submit disputes under the jurisdiction of international arbitration courts, the European Court of Justice (ECJ) loses control over the interpretation and enforcement of ECL, because most awards rendered have the same effects as those of a national court and grounds for appeal are limited (Wehland, 2009). On the grounds of their conflicting and discriminating nature, the EUC has therefore officially requested several EU member-states to terminate their intra-EU BITs (EUC, 2015). Despite this formal request some of these EU member-states, such as the Netherlands and Germany, prefer to maintain their intra-EU BITs (EFC, 2008).

Because of the widespread use of BITs to protect FDI, ample research has been conducted as to the effect of BITs on FDI. As mentioned earlier, Guzman (1997) argued that BITs can be used to overcome the time-inconsistency of government behaviour, removing the unexpected costs and decreases in future returns associated with changing government policies. Fewer policy changes reduce uncertainty, which reduces the risk premium that has to be paid on financing the FDI.

7 Because of the lower cost of financing, BITs can be expected to stimulate FDI inflows. As one of the first empirical studies, Hallward-Driemeier (2003) finds no evidence of any positive effect on FDI flows as a result of BITs to developing countries, and suggests that BITs require good quality institutions to be credible. Neumayer & Spess (2005), however, do find evidence of increased FDI flows to countries with many active BITs, but Tobin & Rose-Ackerman (2005, 2010) find that this increase only occurs when low levels of political risk and good quality institutions are already provided. Opposing these results, Busse et al. (2010) find that BITs do increase FDI flows and may even substitute for poor quality domestic institutions. As can be seen the results in this short summary are not in agreement.

Empirical studies on the effect of intra-EU BITs on FDI have not been conducted to this date, because the discussion in this area mainly focuses on the conflicting legal aspects between BITs and ECL. My research question is “What is the empirical effect of intra-EU BITs on FDI inflows from within the EU?” In an attempt to answer this question I examine the empirical effect of intra-EU BITs on FDI flows between signatory states. I expect that intra-EU BITs have a positive effect on FDI, which might explain why some of the member-states are unwilling to respond to the request of the EUC to abolish intra-EU BITs.

The paper is organized as follows. The first part examines BITs and their potential effect on FDI flows. After this an empirical part follows where a gravity-type model is estimated. After a description of the main results, the conclusion follows.

8

2. BITs and their impact on FDI

In this part the relation between FDI and BITs is examined. The goal is to gain an understanding of the role of BITs in general and the available alternatives. After this the role of intra-EU BITs in particular are discussed.

i.

BITs and FDI

The factors that determine FDI can be divided into three groups. The first are macro-economic factors, such as market size, market development, and endowments in local factors of production. These are the fundamentals of the economy that determine the available investment opportunities and the ability to absorb FDI (Busse et al., 2010). The second group is the institutional quality, which is important for the provision of public goods, such as infrastructure, and to reduce the costs of doing business to promote the well-functioning of domestic markets. Independent legal institutions that enforce property rights and reduce the chance of uncompensated expropriation are also important for potential foreign investors. The third group are FDI related policy factors, such as taxes and regulations on FDI (Bloningen, 2005). It takes long for a government to influence the first two groups, but the third group can be altered more swiftly to create a more favourable FDI environment to attract foreign investors.

However, Guzman (1997) argues that unilateral policy changes to create a more favourable FDI environment are subject to a dynamic time-inconsistency problem, especially in developing countries with poor quality institutions and weak domestic property rights enforcement. In the post-establishment phase the host government can increase its pay-offs by capitalizing on the investment, via policy changes or expropriation, which reduces the investor’s value of the FDI and is a form of political risk. Because of the uncertainty about government behaviour, unilateral regulatory changes without the possibility for investors to enforce them, cannot be expected to lead to more FDI inflows. However, BITs can be used as a binding commitment signal to overcome the time-inconsistency, because its rules and standards can be enforced post-establishment via ISDS, where monetary compensation can be demanded for losses suffered. BITs tie the hands of the current and future governments. This removes the costs associated with unexpected government behaviour, which makes financing FDI cheaper, raises expected profits, and therefore should stimulate FDI inflows (Neumayer & Spess, 2005). Elkins et al. (2006) argue that BITs also increase FDI inflows from non-treaty partners, because the commitment signal boosts overall credibility of the host government in protecting FDI.

Besides voluntarily signing BITs, Elkins et al. (2006) propose that competitive pressures among potential host-countries can also explain the rise in the worldwide number of BITs. They argue that it is rational for individual hosts to sign a BIT in order to gain a reputational advantage

9 over other host-countries that have not signed such a treaty, thereby diverting FDI inflows away from these competing hosts. Elkins et al. (2006) find that host-countries are more likely to sign BITs when competitors have done so in order not to miss out on FDI. Furthermore, the negotiations and ISDS cases cost resources, and when more host-countries sign BITs the reputational advantage becomes less of a distinction. When all FDI in the world is covered by BITs, the reputational advantage can no longer be expected to divert FDI away from competing hosts, and the FDI inflows will be equal to the situation without any BITs worldwide. However, the costs remain in place. Therefore the host-countries face a collective action problem, because they would have been better off if they collectively had not signed any BITs.

In addition, Busse et al. (2010) criticize the one-time game as described by Guzman (1997). They argue that governments are engaged in a repeated game with investors to attract FDI every time-period. A time-inconsistent government, which tries to increase its pay-off by changing FDI policies or expropriation after a FDI has been made is likely to deter future FDI inflows, because these changes affect the current and future attractiveness of the domestic FDI environment. Besides, the time-inconsistent behaviour also harms the overall reputation of the host-country government. Busse et al. (2010) argue that unilateral regulatory changes could increase FDI inflows without the protection of a BIT, because governments care about future FDI inflows and will be careful when it comes to deterring these flows. This carefulness mitigates the time-inconsistency problem, and therefore BITs are not necessary for unilateral regulatory changes to attract FDI.

BITs are a tool for countries to signal imperfectly informed multinationals that the host government is committed to an open and safe investment climate. Poulsen (2010) mentions that there are also alternatives available for investors to mitigate the exposure to political risk. Investors can finance their FDI with host-country credit providers or create a joint venture with a local firm to mitigate political risk stemming from unequal treatment. These strategies create an incentive for the host-country government to treat the foreign investor fairly, because unequal treatment will also affect locals instead. Another alternative is political risk insurance (PRI), which is generally faster in compensation than arbitration. In a survey among PRI providers Poulsen (2010) looked at the relation between BITs and PRI premiums. He found that BITs were of almost no influence in calculating the premiums, which could mean that these PRI firms do not believe that BITs reduce political risk. However, a few PRI firms provide insurance contingent on BITs, where the provider takes over the potential ISDS claims in order to recover the benefits paid out to the investor. The recovery of these benefits with ISDS by only a few PRI firms could indicate that the potential of taking BITs into account as a condition for PRI is not fully realized by the PRI sector (Yackee, 2010). To date ample research has been conducted to the relation between PRI and BITs. It might be interesting to

10 examine whether taking BITs into account affects the profitability of a respective PRI firm, which might be indirect evidence for BITs reducing political risk or not.

One of the most important PRI programs is the Multilateral Investment Guarantee Agency (MIGA), which is part of the World Bank. When FDI is covered by a BIT, MIGA regulations state that there is adequate legal protection to cover the investment with insurance. However, a BIT is not a precondition, so FDI can still be insured when there is no BIT coverage in a host-country. In addition, BITs are just one of the many factors that are taken into account when determining the PRI premium-rates. For example, insuring expropriation risk requires MIGA to take 57 different factors into account to determine the premiums that have to be paid (Poulsen, 2010).

Just like MIGA, other PRI providers generally also have many conditions that have to be fulfilled before insurance is provided. Even if the FDI project is accepted there is often a maximum coverage (Donovan, 2003). Comeaux (1994) argues that investors should always use additional measures besides BITs to mitigate political risk exposure, since the outcome of arbitration is never certain. However, BITs do provide a strong incentive for the host state to honour its obligations, because violating a treaty with another sovereign state can have political consequences for the host-state beyond the treaty itself. In addition, BITs cover all FDI in a host-host-state, sometimes even the FDI made before the treaty was signed. This coverage does not cost the investor insurance premiums and there is also no limit to the claims one can make in an ISDS case.

The extent of legal delegation in the ISDS provisions differs among BITs. The highest degree of delegation is when a BIT allows for the International Centre for the Settlement of Investment Disputes (ICSID) arbitration, which is part of the World Bank and is set up by an international convention, currently signed by 161 states (ICSID, 2016). Part of the convention is that ICSID facilitates a permanent location, rules on proceedings, and awards rendered have the same effects as those of a national court. Grounds for appeal are limited. Prior to ICSID investors had to rely on political protection by the home-state for settling investment disputes with a host-country. ICSID was specifically meant to depoliticize the settlement of investment disputes and to provide a neutral forum where investors could directly enforce treaty provisions outside the often ineffective and biased host-country’s courts (Tietje & Baetens, 2014).

Around 73% of the ICSID caseload to date results from BITs or from ISDS provisions in trade agreements. Another 17% results from individual investment contracts between states and investors. The remainder results from investment laws of host states (ICSID, 2015b). The individual contracts are typically negotiated by very large multinational enterprises (MNE) only. ICSID is used for investor-state dispute settlement more often than all other available settlement bodies combined (Gaukrodger & Gordon, 2012). The alternatives require both parties to agree on many of the areas

11 already provided by ICSID, such as a location, rules, and award procurement, which takes time and is therefore generally less efficient than ICSID (Allee & Peinhardt, 2010).

In short, BITs are a way to overcome the time-inconsistency of government behaviour. BITs can serve as a signal of commitment to a favourable investment environment. There are alternatives available for investors, but BITs have the broadest coverage and this coverage does not require insurance premiums to be paid.

ii.

Intra-EU BITs

The reason why EU member-states concluded BITs in the absence of a community regime is because multilateral negotiations in the field of FDI protection were not successful in the past. Developing countries campaigned aggressively against multilateral FDI rules and therefore international customary law offers little protection for foreign investors. To compensate for the lack of FDI protection in international law, the capital exporting EU member-states started to negotiate treaties on the bilateral level with non-EU countries (Elkins et al., 2006). Strengthening investor protection and the ability to use arbitration to settle investment disputes were also the main reasons why the EU member-states signed BITs with the CEEC before they joined the EU (EUC, 2015). In figure 2 around 75% of the currently active intra-EU BITs include a reference to ICSID for dispute resolution, and therefore provide the highest degree of delegation. In total 36 of the intra-EU BITs have been terminated to date.

Figure 2: Number of intra-EU BITs per EU member. Source: UNCTAD, 2016.

12 Figure 3: EU countries with the number of active intra-EU BITs. Source: UNCTAD, 2016.

Figure 3 shows that the CEEC are involved in intra-EU BITs relatively more often than the rest of the EU member-states. Now that the CEEC have joined the EU, the EUC is of the opinion that the extra protection provided by the intra-EU BITs is no longer necessary in the single market, which is protected by the ECL. The EUC argues that the 36 intra-EU BITs that have been terminated in mutual consent are proof of the fact that they are obsolete (EUC, 2015). ECL and the intra-EU BITs both share the same goal of promoting FDI and to create a stable legal framework regarding FDI. However, according to Eilmansberger (2009) the main purpose of the intra-EU BITs is to provide protection after the FDI has been made, which it does by its rules and standards on the admission and treatment of investors, and by prohibiting expropriation. In addition, BITs provide access to ISDS to enforce these provisions. The main purpose of ECL is to eliminate obstacles for cross border capital movements within the EU and ECL only provides basic outlines for the protection of FDI. Therefore BITs concentrate on the post-establishment phase, while ECL concentrates on the pre-establishment phase.

13 When it comes to expropriation there are two different forms. The first is direct expropriation, where the host government takes ownership of the FDI. The second form is indirect, where government measures or regulation disrupt the FDI to the point of non-functionality. The former seldom takes place in modern times, but the latter is prevalent (Tietje & Baetens, 2014). When it comes to indirect expropriation monetary compensation is rare in national rulings of the major European countries according to Van Aaken (2009), because the legal doctrines are focused on primary remedies, such as annulling or prohibiting the government actions impairing the investor. When a ruling annuls a government measure, monetary compensation can be sought as a remedy of last resort. Arbitration does not require that all local remedies are exhausted before a foreign investor can get monetary compensation, instead compensation can be demanded directly. Therefore using arbitration to get compensation is more efficient than using most local remedies. In addition, international law in general does not always allow a tribunal to annul a national government’s measure, making it harder to enforce a primary remedy rather than a monetary reward (Allee & Peinhardt, 2010).

According to Dimopoulos (2011a) expropriation is not prohibited by ECL and compensation is only required for illegal expropriation. However, ECL does not determine the conditions under which expropriation of foreign investments is illegal and to what extent compensation is required. This still is an exclusive competence of the EU member-states, therefore BITs provide better and unconditional protection when it comes to expropriation.

Besides the possibility to make direct monetary claims in arbitration cases there are other advantages. Firstly, arbitration transfers the interpretation of the treaty to an independent international tribunal, instead of the potentially biased host-country’s domestic courts. Arbitration depoliticizes investment disputes between states and investors, avoiding political lobbying, delays, and corruption. Also, the investor has a vote in the appointment of the arbitrators in the tribunal, which is generally not the case in domestic courts. These arbitrators are typically experts in their respective field, where domestic judges on the other hand may lack the time and resources to acquire the expertise necessary to interpret the technical fields of the topic at hand (Tietje & Baetens, 2014). In addition, resolution via international arbitration is generally faster and cheaper, because the legal institutions of lesser developed EU host-countries, such as Slovenia, Slovakia and Poland, are circumvented. This is even the case for more developed economies such as Italy and Spain, where official procedures around investment dispute can take up to two years (Van Aaken, 2009).

Apart from the advantages there are also unresolved issues with BITs, such as the extent to which a state is required to compensate a foreign investor for regulations that benefit the public interest, such as safety or environmental regulations, which could be interpreted as impairing the foreign investor. In this way BITs could even lead to regulatory chill, where governments refrain from

14 implementing policies that could benefit public interest, because of potential claims (Gaukrodger & Gordon, 2012). Examples of intra-EU ISDS cases based on regulations beneficial to the European public are numerous, such as the major Swedish energy supplier Vattenfall suing the German government based on legislation to phase out nuclear and coal-fired power plants, which is part of the German transition to a more sustainable and environmental friendly energy sector (UNCTAD, 2016b).

Investors might also abuse the ISDS provisions by trying to obtain compensation for FDI that turned out to be unprofitable due to normal market conditions or a bad gamble. Any government policy can potentially be challenged as long as it affects the FDI, even if this policy is not the primary cause for the FDI turning unprofitable (Yackee, 2010). The potential for compensation might even lead to moral hazard and adverse selection, where investors might work less hard to turn the FDI into a success, or where investors are attracted to states with the best potential of successful arbitration cases (Neumayer & Spess, 2005). Hallward-Driemeier (2003) argues that with the rising number of successful arbitration cases more investors will start looking at ways to exploit the terms of the treaties, seeking compensation for risks that they had not thought of in the absence of the treaty. These might also be potential problems for the intra-EU BITs, where domestic law changes to comply with ECL have already led to conflicts with BIT provisions. An example of an ISDS case based on regulatory changes to comply with ECL is the Belgian firm Electrabel suing the Hungarian government, because under ECL the regulation in question was marked as unlawful state aid, benefitting Electrabel (UNCTAD, 2016b). However, the extent to which investors exploit the treaties and whether regulatory chill actually occurs is hard to measure in an objective way and has not been proven to date.

In an effort to measure how acquainted European investors are with BITs the EUC (2000) conducted a survey. The goal was to obtain an understanding of the problems encountered by European enterprises when engaging in FDI. This survey shows that only 50% of the 300 respondents heard of BITs and only 10% had working knowledge. This 10% includes companies that considered BITs when they were facing problems involving the host government regarding their FDI, but also the companies that considered BITs as entry condition. This can be seen as indirect evidence that BITs probably do not lead to more FDI inflows from within the EU, because few companies are actually aware of BITs, and even fewer consider them as entry condition. However, the growing number of ISDS disputes in the world might indicate that this awareness is increasing among investors in general, which could lead to BITs playing a larger role in current and future investment decisions (Poulsen, 2010). On the other hand, Tietje & Baetens (2014) argue that the number of disputes is also correlated with the FDI stock in general, so the growing number of disputes might also just be a consequence of the increased FDI stock in the world.

15 The growing number of arbitration cases and their awards rendered show how successfully investors have protected their interest via BITs (Dimopoulos, 2014). As can be seen in figure 4, there are 696 known ISDS disputes to date, of which more than half have been initiated by investors from one of the EU member-states. The most active initiating investors in the EU originate from France, Germany, the Netherlands, and the United Kingdom. From these four countries a total of 228 disputes were initiated, which is a third of the world total. In total 69 of these disputes were initiated against other EU member-state governments, which is around 40% of the total number of disputes initiated against all EU governments.

Figure 4: Known ISDS cases to date. Source: UNCTAD 2016b

Almost 80% of all disputes where one of the EU member-state governments is sued originate from within the EU. Perhaps the most striking fact is that the governments of the four countries where most disputes originate from are only responding to five disputes in total from the entire world, even though all BITs are reciprocal in nature. The fact that so many disputes originate from these four countries, while the governments themselves are hardly sued might be a reason why these countries

16 are such strong proponents of maintaining their intra-EU BITs. The figure that 19% of the world total of disputes is intra-EU might also indicate that intra-EU BITs still offer relevant protection, even with the ECL in place. On the other hand, this could also indicate that investors from the four most active countries found ways to exploit the intra-EU BITs, because the new EU members probably have to make more regulatory changes to comply with ECL after their entry.

The Czech Republic, Hungary, Poland, and Spain are the EU member-states which are known to be in favour of terminating their intra-EU BITs (EFC, 2008). The governments of these countries faced a total of 80 disputes from within the EU, which is around 61% of the total. These uneven numbers support the argument of the EUC that intra-EU BITs discriminate among investors in the EU. However, within the EU Dutch BITs are most often used for initiating disputes, but 75% of these disputes are made by investors that have no ties to the Netherlands whatsoever. This is a phenomenon called treaty shopping, where investors from outside the Netherlands use shell corporations in order to benefit from the extensive Dutch BIT network. If these foreign investors originate from other EU member-states, this diminishes the discrimination on grounds of nationality (Gaukrodger & Gordon, 2012).

As a final remark, Lavranos (2010) argues that the EUC has not gained its exclusive competence on FDI to lower overall standards of protection for European investors. Therefore the extra benefits enjoyed by investors under intra-EU BITs should be extended to the other European investors instead of terminating the treaties. This would also end the discrimination.

In short, the intra-EU BITs were initially signed to provide protection for FDI in the CEEC in a period in which they had not yet acceded to the EU. Even though BITs seem to provide better protection than ECL in cases of expropriation, there is also the possibility of regulatory chill, and of investors exploiting the treaties. The countries from which most intra-EU disputes originate are proponents of maintaining the treaties. These countries do hardly receive any claims themselves, while the four most outspoken opponents of the intra-EU BITs receive more than half.

17

3. Empirical application

Here the most relevant empirical literature with respect to BITs is described first. After this, the methodology used in this paper is described based on the discussed literature. This part is concluded by a discussion of the estimation results.

i.

Literature:

The gravity model is heavily used in the empirical trade literature to explain country-pairs’ trade flows and the basic equation specifies that trade flows between a country-pair are a function of the GDP of each country and the distance between the two countries. By applying the literature on the determinants of horizontal and vertical FDI by MNEs to the gravity model, the model can also be used to explain variation in country-pairs’ FDI flows. Horizontal FDI, which are investments to be able to produce in and for the host-country’s domestic market, is motivated by host-country’s market demand and by the presence of barriers to trade. The former is often measured by GDP and by the growth rate of GDP as an indicator for future market development, and the latter by the host-country’s openness to trade. Vertical FDI is motivated by cost savings and the goods produced are often exported from the host-country to other markets. The former is often proxied by differences in wage levels. For the exports transportation costs are important, which are often measured by distance (Blonigen, 2005). The application of the gravity model on FDI flows would then also be useful to investigate the effect of BITs on FDI, which is done in all papers discussed below.

In almost all these papers fixed-effects are used in the gravity model specification in order to control for unobserved and missing variables, which otherwise result in omitted variable bias. If this bias is present, the coefficients of the explanatory variables in the model will be estimated incorrectly, because they also include part of the correlation which actually belongs to the missing variables. The fixed-effects are used to capture the effect of these missing variables in order to be able to estimate the other explanatory variables in the regression correctly. Two different fixed-effects are used. The first are time fixed-fixed-effects, which controls for factors that are constant across individuals but vary over time, such as the trend increases in worldwide FDI. These controls can be added to the model-equation in the form of a dummy for all but one year. The second are country-pair fixed-effects to control for factors that are specific to the country-country-pair and remain constant over time, such as the distance or historical and other ties between the two countries. These effects can be added to the gravity equation in the form of a dummy for all but one country-pair. Including a dummy for every year and country-pair results in perfect multicollinearity, meaning that there is a perfect linear combination of the regressors. This is also called the dummy variable trap and prevents the estimation of the regression (Stock & Watson, 2012).

18 In the papers discussed below two different approaches can be distinguished. The first is a bilateral approach, where the FDI outflow from a source-country to a host-country is used as dependent variable. Outflows are used because the OECD is the main data-source for bilateral FDI flows, which only provides data for its members. For these OECD countries outflows to the rest of the world are reported, and by using these outflows countries that do not report their FDI inflows can be included in the data-set. In the bilateral approach the effect of a country-pair’s BIT on their respective bilateral FDI flows is studied by using a dummy, which takes the value of one if the BIT between the country-pair is active and zero otherwise. The second approach is an aggregated approach, where the aggregated country-level FDI inflow is the dependent variable. The effect of a host-country’s total number of BITs on its aggregated FDI inflow is then studied. The two approaches use different data and account for BITs in a different way.

The first paper to empirically study the effect of active BITs on FDI flows to developing countries is Hallward-Driemeier (2003). She estimates a fixed-effects gravity model using the bilateral approach with 537 country-pairs for the sample-period of 1980 until 2000. Besides the basic gravity model variables of host-country size and openness to trade, she uses host-country GDP per capita as an additional proxy for horizontal FDI’s potential market demand, source-country GDP as an indicator for the available pool of FDI, inflation as a proxy for macro-economic stability, and the gap in the average years of education between the country-pair to proxy a cheaper low skilled labour force for vertical FDI. As a proxy for political risk she uses the index of law, order, and corruption from the International Country Risk Guide (ICRG). This index is also often purchased by MNEs to assess country risk levels.

Hallward-Driemeier (2003) uses two different specifications of her dependent variable, first the net outflow from the source-country to the host-country, and secondly the ratio of the net FDI flows from the source-country to the host-country and the total FDI outflows from the respective source-country. Using the ratio is a way to control for the upward trend of both worldwide and individual FDI, because the ratio simply transforms the series so that the trend is removed. This trend might otherwise cause spurious significant relationships in the model-estimation.

Hallward-Driemeier (2003) argues that reverse causality between her BIT variable and FDI might be a problem in her model, which means that causality not only runs from the regressor to the dependent variable, but also from the dependent variable to the regressor. The regressor then is not exogenous, which is one of the assumptions of ordinary least squares (OLS). This endogeneity results in biased and inconsistent OLS estimates, because the regressor is correlated with the error term and therefore picks up unwanted effects in the estimation. As a solution for this instrumental variable (IV) regression can be used, where the problematic regressor is replaced by an exogenous variable that is correlated with the regressor, but not with the dependent variable (Stock & Watson, 2012).

19 Hallward-Driemeier (2003) recognizes that this feedback might exist if ratifying a BIT increases FDI flows to a developing country, but investors on the other hand also lobby in their home-country to sign a BIT with the host-country in which they are already involved to protect their current and future investments. Hence, the causality works both ways. As a solution for this she uses the number of BITs that a host-country has signed with other partner-countries as an instrument for the BIT signed between a country-pair.

In her main model specification the coefficient of the BIT dummy is negative and insignificant, while she expected a positive value. In addition, she examines the effect of BITs in the years after and before its ratification, which does not result in any significant results for the BIT variable either. Finally she includes an interaction term between her BIT variable and her political risk proxy in her model. The coefficient of the interaction term turns out to be significantly positive in combination with a negative BIT coefficient. A higher value of the ICRG index represents lower political risk. The positive interaction term therefore suggests that BITs are more effective in countries with better quality institutions and lower political risk. However, this is where they are least necessary, which suggests that BITs are complements to good institutional quality.

Tobin & Rose-Ackerman (2005) use the aggregated approach, because they argue that the bilateral approach might lead to an underestimation of the effect of BITs on FDI, because the signalling function towards non-treaty partner investors as proposed by Elkins et al. (2006) is not captured. In addition, much more country level FDI data is available compared to bilateral flows data, which allows for a more representative sample of 63 developing countries, instead of the 31 used by Hallward-Driemeier (2003). Just like Hallward-Driemeier (2003), Tobin & Rose-Ackerman (2005) examine the same period and use the ratio of the country-level FDI inflow in a host-country divided by the total FDI inflows to all developing host-countries as dependent variable.

As the explanatory variables the logarithms of GDP, growth of GDP and population are used as a proxy for the host-country’s market size. Logarithms reduce the spread of the data, meaning that the estimation is less sensitive to outliers, which might drive the results of Hallward-Driemeier (2003). In addition, the logarithm of the cumulative number of BITs signed by a particular host-country is used to account for the signalling effect towards investors from non-treaty partner-countries. The logarithm takes account of decreasing returns to scale of signing additional treaties. For political risk a different broader index from the ICRG is used as a proxy for the entire host-country investment environment, which includes the index of Hallward-Driemeier (2003) as well as other institutional and socio-economic indicators. The exclude openness, because it is not independent of FDI. This is a strange choice, because they could have used an instrument, as they do for their BIT variable.

20 Tobin & Rose-Ackerman (2005) take five-year averages of all variables except BITs to smooth the highly volatile year-to-year fluctuations. The reason for not taking the average of the BIT variable is because according to them it is unlikely that the signing of such a treaty has an immediate effect on FDI. To account for the potential reverse causality between BITs and FDI two instruments are used. The first is the level of democracy in a host-country, because source-countries tend to be democracies and tend to sign treaties with other democracies. The second is a time variable, because governments learn about the potential benefits of BITs over time, which increases their incentive to sign a BIT.

Tobin & Rose-Ackerman (2005) find a negative but insignificant effect of BITs on FDI. When they include an interaction term between political risk and the number of signed BITs, the coefficient is significantly positive. A higher value of their broader ICRG index also represents lower political risk, so the negative coefficient suggests that BITs are complements to a good investment environment. They argue that countries need domestic institutions in place to enforce property rights and reduce political risk, which make BITs credible and valuable for investors.

Neumayer & Spess (2005) criticize the findings of Tobin & Rose-Ackerman (2003), because their political risk measure also includes socio-economic indicators such as religious tensions and unemployment, which have no direct relation with the quality of institutions or with property rights. Therefore they are not relevant in order to study the effect of BITs. Neumayer & Spess (2005) also criticize the use of a sample of 63 countries, while much more data is available. For instance, Tobin & Rose-Ackerman (2005) exclude transition countries like China and all CEEC countries because of data limitations, which could lead to selection bias. To include as many observations as possible Neumayer & Spess (2005) investigates a longer sample period from 1970 until 2001, and a sample of 119 developing and transition host-countries.

In order to test the effect of the measure of risk five different commonly used measures are used in separate regressions. Four of these are subsamples of the ICRG index, and the fifth is the index developed by Henisz (2000). Neumayer & Spess (2005) use the cumulative number of signed BITs because the log of zero BITs is not defined. How Tobin & Rose-Ackerman (2005) handle the log of zero BITs is not explained. Another difference with the previous papers is that Neumayer & Spess (2005) use the logarithm of the ratio of FDI inflows in a particular host-country divided by the total FDI inflows in all their host-countries. They do this to reduce the spread of the data and to improve their model fit. In order to do this negative and zero values are replaced with a value of one, because the logarithm is only defined for values larger than zero. This is a peculiar choice, because with this transformation years of disinvestments with negative FDI flows are no longer taken into account, which introduces a positive bias to their results. The rest of the explanatory variables are equal to Tobin & Rose-Ackerman (2005).

21 Neumayer & Spess (2005) reject the use of instruments, because of the reason that every explanatory variable used is potentially endogenous, and it is not possible to find appropriate instruments for each variable. For instance, a country’s level of income or economic growth might attract FDI, but both variables may also be a consequence of FDI. As a solution all explanatory variables are lagged by one year, in order to solve the feedback of the endogenous variables.

The results show that countries that have signed more BITs attract significantly more FDI. Neumayer & Spess (2005) also include an interaction term between their BIT variable and political risk proxies. When they use the ICRG index, they find a significant negative coefficient for the interaction term, meaning that the effect of BITs on FDI decreases with lower political risk in a country. This would suggest that BITs are substitutes to poor institutional quality and high political risk. These results do not change when instead the same sample of countries and sample period as Tobin & Rose-Ackerman (2005) are used, which can be seen as evidence that the differences in political risk measures are important. On the other hand, it could also indicate that the transformation of negative and zero values to one leads to positively biased results, or that the lagging of explanatory variables is not as thorough as using IV.

According to Busse et al. (2010) lagging the explanatory variables by one period does not completely remove reverse causality, it simply changes the channel of the feedback by one period. Therefore Busse et al. argue that the results of Neumayer & Spess (2005) are still biased by endogeneity.

In addition, none of the aforementioned papers control for unilateral regulatory changes, because of the assumption that these changes suffer from a time-inconsistency problem and should therefore not lead to more FDI. However, Busse et al. (2010) argue that a time-inconsistent government is likely to deter future FDI inflows once a repeated game between foreign investors and the government is taken into account, as described in more detail in section 2.i. Therefore, unilateral regulatory changes to improve the domestic FDI environment may also stimulate FDI inflows without the protection of a BIT. To control for these changes Busse et al. (2010) include a proxy for unilateral capital account liberalization in their model. When this control is taken into account the signalling effect towards non-treaty partners cannot be entirely attributed to BITs, because these unilateral regulatory changes have occurred at the same time. In addition, the signalling effect of BITs is no more credible than the unilateral regulatory changes and only the extra protection offered to investors from signatory countries matters. Therefore Busse et al. (2010) use a dummy for an active BIT, because only when the treaty is in force it offers more protection. In order to examine the effect of this extra protection and to separate the FDI flows stemming from treaty and non-treaty partners, Busse et al. (2010) use the bilateral approach.

22 Busse et al. (2010) use a sample of 2313 country-pairs for the period 1978 until 2004. In addition, they take three-year averages of all the variables to reduce the bias of volatile year-to-year variation. It is also the first paper to run a regression on a sample that includes developing countries as potential source-countries. The aforementioned papers all exclude developing source-countries and focus on developed source-countries instead. The exclusion of developing source-countries is outdated, because in recent years FDI flows originating from developing countries started to increase. The disadvantage of including developing countries as sources is that there are many missing values and zero observations. The negative and missing values are set equal to zero, because missing values are most likely to be zero values, except from the confidential values which are not reported.

As dependent variable Busse et al. (2010) use the ratio of FDI outflows from a source-country to a country, divided by the total FDI outflows from that particular source-country to all host-countries. Like the aforementioned papers Busse et al. (2010) use GDP, the growth of GDP, and inflation as explanatory variables. The difference in GDP per capita between the source and host-country is used as a proxy for cost savings for vertical FDI, because it is an indication for lower incomes in the host-country. The dependent variable and the four explanatory variables are transformed according to the following logarithmic transformation formula to reduce the effect of outliers in the data. Busse et al. (2010) do not provide any arguments for why they use this specific transformation, besides the argument to still be able to use the zero values. However, to be able to use zero values plenty of other transformations could also have been used.

Other included variables are the Henisz (2000) index of political constraints as a proxy for political risk, a dummy for double taxation treaties, because these treaties lower effective tax rates and should therefore stimulate FDI, a common currency dummy, which results in lower transaction costs and should therefore increase FDI flows, and a dummy for free trade areas and customs unions.

In order to overcome the possible endogeneity of BITs and GDP Busse et al. (2010) the Generalized Methods of Moments (GMM) system estimator. This method was described by Blundell & Bond (1998) for a setting with a small number of time-periods and many individual countries. It can be used with panel data with endogenous explanatory variables, and in the absence of any strictly exogenous explanatory variables or instruments. Strict exogeneity rules out any feedback from current or past shocks to current values of the variable, which is necessary for IV regressions.

The GMM-system estimator works by simultaneously estimating a levels equation and a first-difference equation for each year of the sample. A first-first-difference equation can be created by subtracting the value of the previous period from each value of the current period. In this matrix, or system of equations, moment conditions are cleverly used to derive a set of valid instruments for the

23 endogenous variables. Moments are measures that describe the shape of the probability distribution, such as the mean and variance. The instruments are the lagged values of the variables from the two estimated equations, where the lagged values from the levels regression are used as instruments in the first-difference equation, and the lagged values from the first-difference equation are used as instruments in the levels equation. For each equation the instrument used is allowed to differ, which is useful because after each period more lagged values of the instruments are available. The parameters of the estimated model are the ones that provide the best fit to the different moment conditions, where each parameter sets a sample moment to zero. These moment conditions typically cannot all be satisfied at the same time, therefore a quadratic objective function is minimized to make a trade off between satisfying the different moment conditions. This is similar to the way IV estimation works, but the GMM-system estimator is more efficient in the system setting described above, which means that there is a higher probability that the estimates are closer to their true value (Stock & Watson, 2012)

The main finding of Busse et al. (2010) is that BITs significantly promote FDI flows to developing countries. In addition, they also include an interaction term between their political risk proxy and their BIT dummy, which is significantly negative. A higher value for their political risk proxy actually means lower political risk. A negative correlation therefore means that BITs are more effective in a host-country with higher political risks. This can be seen as evidence for BITs as substitutes for host-country domestic institutions and political risk.

In a later work Tobin & Rose-Ackerman (2010) also use the GMM-system estimator with a similar dependent variable and explanatory variables as Tobin & Rose-Ackerman (2005). In this later work the following logarithmic transformation for the dependent variable is used to reduce the effects of outliers in the data:

Their main argument for this transformation is that it allows them to preserve the information of negative and zero observations in their dataset. Tobin & Rose-Ackerman (2010) argue that replacing the negative, missing, and zero values, as is done by Neumayer & Spess (2005) and Busse et al. (2010), might lead to biased results. However, there are many other transformations available to be able to use negative and zero observations, and as such, this transformation itself might also lead to biased results.

Tobin & Rose-Ackerman (2010) find that the effect of a BIT on FDI inflows is negative and insignificant. When the interaction between an individual country’s BITs and the total number of BITs in the world is included, the BIT coefficient turns positive, but remains insignificant. The interaction

24 term is significantly negative, suggesting that any positive effect of signing a BIT for an individual country decreases when more BITs are signed in the world.

The last paper discussed is written by Bevan & Estrin (2004), in which the determinants of FDI to European transition countries are studied by using a panel-dataset of bilateral FDI from 1994 until 2000. Besides the gravity variables GDP, openness, and distance, the difference in unit labour costs between a particular host and source-country is also included as a potential driver for vertical FDI. Interesting to notice is that Bevan & Estrin (2004) use a random-effects estimation for their main model, instead of the fixed-effects estimations used by the other papers. They do this based on an insignificant Hausman test-statistic. The Hausman test can be used to test whether the unobserved time-invariant elements are correlated with the explanatory variables used. If the test-statistic turns out to be insignificant, a random-effects model is preferred, which means that the dummies for each country-pair are not included in the regression.

Bevan & Estrin (2004) find that the market size, proximity to source-countries and unit labour costs have a significant effect on FDI, but their political risk variable is insignificant. As an explanation for this they argue that the announcement of entry negotiations to the EU may reduce perceived country risks, because meeting the requirements that are necessary for joining the EU require adaptation of EU legislation and a certain level of quality of the economic management and institutional development. Besides this Bevan & Estrin (2004) argue that investors tried to get a foothold in the potentially profitable future markets within the predictable and secure institutional settings of the EU. This is why the announcement of entry negotiations increased FDI inflows. Bevan & Estrin (2004) do not include BITs in their study, but the insignificance of political risk might indicate that the extra protection of intra-EU BITs are not important for attracting FDI.

In short, there are two main approaches, the bilateral and the aggregated flows approach. The results of this empirical literature discussion show that the estimation results as to the role of BITs are not in agreement, which might be due to the different estimation techniques used, or due to the use of different explanatory variables and approaches. Other differences that might have an effect on the results are the specification of the dependent variable, the years of averaging the data, and how the authors deal with negative and zero values in their studies.

ii.

Methodology

One of the benefits of the aggregated approach is that more data is available, but this method would requires the use of country-level aggregated FDI inflows from the entire world. On the other hand, the bilateral approach method allows for a better analysis of the specific effect of BITs on FDI flows from a partner EU-country. This method would allow me to examine the bilateral FDI flows within the EU, and to exclude the FDI inflows from the rest of the world. Since I am interested in the effect

25 of intra-EU BITs on FDI flows within the EU, the bilateral approach seems more appropriate, and therefore I use this approach for my thesis. As country sample I use the EU countries specified in figure 5. To include as many observations as possible I use all available data from 1986 until 2012.

Figure 5: Country sample.

From the literature discussion in section 3.i it becomes clear that the authors use different specifications as dependent variables and different time-spans of averaging their data. I expect that both the dependent variable specification and the time-spans for averaging my data would have a great impact on the results. Therefore, in order to examine the effect of these choices I include six different specifications of the dependent variable, shown in figure below, and four different time-spans of averaging my data, from one through four years. This also allows me to test the robustness of the specifications across different time-spans and to find out which specification performs best in line with the theory discussed. After this I will examine the best performing specification in more detail.

26 Figure 6: Different dependent variable specifications.

In order to allow for years of disinvestment I do not transform the missing values and negative values to zero or one. In the case of zero values for specification II and IV, the zero value is used to prevent missing values from an undefined logarithm. In the case of negative values for specifications II, and IV through VI, the logarithmic transformation is done for the absolute value of the observation. After the transformation the observation is multiplied by negative one.

Specification I represents the untransformed bilateral FDI outflow from source-country i towards host-country j. This specification is most sensitive to the effects of outliers in the data, which could drive the results. To reduce the effect of outliers in the data a logarithm can be used, as is done in specifications II, IV, V ,and VI. In addition, the use of a logarithm as dependent variable can be useful because the coefficients of the explanatory variables will then represent elasticities. This means that if any of the explanatory variables changes by one unit, the dependent variable is expected to change by 100*β percent

Another method to reduce the effect of outliers is to use the ratio of bilateral FDI flows and total outflows of the respective source-country, such as specifications III through VI. This also reduces the probability of a unit root, which is a potential trend driving some of the variables, such as the upward trend in worldwide FDI flows potentially driving the individual country’s FDI inflows. A ratio can also be used to model the relative attractiveness of a host-country with respect to the total outflows of a source-country. In specification IV through VI a logarithm is combined with using ratios to reduce the effects of heteroskedasticity even further. The six different specifications are used as a dependent variable for the model below.

27 β β β β β β β β β β β β β β β β β

The subscript i represents the source-country, j represents the host-country, and t the time-period. The first term is the lagged dependent variable ( ), which is included because current

levels of FDI are likely to depend on past FDI flows in the form of up investments. The follow-up investments could be smaller than the initial investments, in which case there is a negative correlation between current and past FDI flows, or larger than the initial investment, in which case there is a positive correlation.

The host-country’s GDP (LogGDPHost) and GDP growth (Growth) are expected to have a positive effect on FDI inflows, because they are proxies for potential market demand for horizontal investments. The source-country’s GDP (LogGDPSource) is also expected to have a positive effect on FDI, because if a country has a higher income its investors can engage in relatively more FDI. Inflation (Inflation) is a proxy for macro-economic stability. A higher level of inflation is expected to have a negative effect on FDI. Openness to trade (Openness), specified as the sum of imports and exports divided by GDP, can have a positive effect on FDI, because more openness stimulates export driven vertical FDI. On the other hand, it can also have a negative effect, because the incentive for market driven horizontal FDI will be larger in a closed economy. So the sign of the variable depends on which type of FDI dominates.

The difference in GDP per capita between the host and source-country (LogDiff) can have a positive effect on vertical FDI, since a larger difference could indicate that the level of income in the host-country is lower. On the other hand, a larger difference might also deter horizontal FDI, because the potential market demand is smaller. Therefore the effect of LogDiff on FDI depends on the dominating type of FDI. The use of logarithms for LogGDPHost, LogGDPSource, and LogDiff is done to reduce the effect of heteroskedasticity.

For the BIT (BIT) variable I use the year in which an intra-EU BIT between a country-pair enters into force, because only an active treaty offers protection. In addition, Busse et al. (2010) argue that the signalling function is not attributable to the BIT itself, but to the unilateral changes made to the investment climate by a respective host-country. Besides this, not all signed intra-EU BITs have actually been ratified, hence they do not offer additional protection and should have no effect on FDI. The variable PolConxBIT is an interaction term between the political risk proxy and intra-EU BITs. According to Tobin & Rose-Ackerman (2005) political risk in a host-country needs to be

28 reduced to a certain level for BITs to become a credible tool to attract FDI. However, Busse et al. (2010) argue that BITs can substitute for poor institutional quality and high levels of political risk. The inclusion of the interaction term allows me to examine whether intra-EU BITs are complement or substitute for institutional quality. A negative coefficient for the interaction term and a positive coefficient for the BIT variable would suggest that BITs substitute for institutional quality. On the other hand, a positive interaction term and negative coefficient for the BIT variable would suggest that BITs complement institutional quality. Therefore the effect of the BIT and PolConxBIT variables can either be positive or negative.

In addition, I use dummies for membership of the EU (EUMem) and whether the countries share a common currency (CommCurr). These are both expected to have a positive influence on FDI. Membership of the EU means that the country falls under the ECL, which has the focus of stimulating FDI in the EU and a common currency reduces transaction costs between two countries. A double taxation treaty (TaxTreat) could reduce the effective tax base and makes filing taxes easier for a MNE, which should stimulate FDI. On the other hand, Hallward-Driemeier (2003) mentions that these treaties could also be used to reduce tax evasion, which could have a negative effect on FDI because potential gains of tax evasion might be lower. So the effect of TaxTreat could either be positive or negative.

In line with the discussed literature I use a proxy for the political risk of the host-country (PolCon). Two different measures are used in the previously discussed papers, which I would ideally like to include both in separate regressions to see their respective effect in my model. However, the ICRG data are not openly available. Therefore I only use the index for political constraints developed by Henisz (2000), which is also used by Bevan & Estrin (2004), Busse et al. (2010), and Neumayer & Spess (2005).

The Henisz index was designed to measure the likelihood of changes in the government’s policy regime in an objective way. This is done with a spatial model of political interaction to derive the extent to which the current political actors and their replacements are constrained in their choice to alter the policy regime. For this the spatial model uses the number of independent branches of the government with veto power over policy changes, based on data of party composition of these branches. More alignment over different branches increases the probability of policy changes. Scores in this index starts from zero, which indicates that the government has complete political discretion, meaning that existing policies can be changed instantly at all time. The exact opposite of this situation is when policies cannot be changed at all, which receives the value of one. All values are less than one since this situation does not occur in reality, because policies can always be changed when an agreement is reached. The measure has a strong correlation with the ICRG index (Henisz,

29 2000). The effect of PolCon is expected to have a positive effect on FDI flows, since a higher value represents decreases in the political risks of changing government behaviour.

To account for unilateral regulatory changes this paper uses the Chinn-Ito index (KaOpen), as is done by Busse et al. (2010). This index measures the capital account openness by looking at the intensity of restrictions on cross-border financial transactions. It is based on data reported in the International Monetary Fund’s Annual Report on Exchange Arrangements and Exchange Restrictions. The Chinn-Ito index ranges from zero to one, where a higher value means that a country is relatively more open to cross border capital (Chinn & Ito, 2007). For this reason a positive link between KaOpen and FDI is expected.

The final variables in the model are Year, which are dummies for every time-period, and FixedEffects, which are dummies for each country-pair. The inclusion of the year-dummies is done to control for common trends over time, which are the time fixed-effects as explained in section 3.i. The country-pair dummies are used to control for omitted variable bias due to time-invariant unobservable elements, which are the country-pair fixed effects, also discussed in section 3.i.

At first I estimate the different regressions by using fixed-effects OLS, in line with most of the aforementioned papers. I use fixed-effects because it is very likely that there are time-invariant unobserved elements in my country-sample, such as the current and historical ties between the countries. For the OLS estimations it is necessary to assume that the explanatory variables are all exogenous.

However, from the aforementioned papers it follows that some of the explanatory variables might be endogenous. Hallward-Driemeier (2003) argues that BITs might cause FDI inflows, but could also be a consequence of it. In addition, Neumayer & Spess (2005) argue that this feedback might exist for all the common explanatory variables used in the gravity model. Because it is not possible to find appropriate instruments for all these explanatory variables Neumayer & Spess (2005) lag their right-hand side variables by one period. However, Busse et al. (2010) argue that these one-period lags simply change the channel of feedback by one period. Therefore they use the GMM-system estimator instead, as explained in section 3.i. The GMM-system estimator was designed for a setting with a small number of time-periods and many individuals. It can be used for panel-data with possible endogenous explanatory variables, in the absence of any strictly exogenous explanatory variables or instruments.

Almost all of my explanatory variables could potentially be endogenous, such as BITs, LogGDPHost, and Polcon. For example, high political risk in a country might be a reason to sign a BIT, but signing a BIT could also reduce the political risk. Since it is not possible to find instruments for all explanatory variables, I use the GMM-system estimator in my final regressions to see whether the results differ significantly from the OLS estimations.