Foreign capital flows and BITs – Why the effect of
BITs on the directional flow of capital has shifted
Master Thesis MSc Finance (Corporate Finance)
By S.P. (Sake) Willems
11878827
Abstract − Over the past decades, bilateral investments treaties (BITs) have become one of
the most widely used legal instruments for protecting foreign investments. However, whether and how BITs stimulate foreign capital flows is still open to debate. I use two large
comparable datasets of dyadic foreign direct investment (FDI) and cross-border M&A to assess the effect of BITs on foreign capital flows. Contrary to prior research, I observe a significant increase in FDI flows between two developing countries. This result supports the view that firms in developing countries, there being subject to poor governance, have
developed skills to secure their outward FDI to other developing countries better. There is no noticeable effect for M&A activity, indicating that these measures of foreign capital flows cannot be interchanged one-to-one. Subsequently, I analyze the impact of BIT ratification on foreign capital flows, conditional on institutional quality at the capital receiving country. Using multivariate fixed-effects regressions, I find limited evidence that BITs can substitute for high institutional quality. Understanding the implications of entering into a BIT remains a challenging task for policymakers, in particular those from developing countries.
Supervisor: dhr. dr. S.R. (Stefan) Arping
December 2018
2
Statement of originality
This document is written by Student Sake Willems 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
Table of Contents
1. Introduction 4
2. Theoretical Framework 7
2.1 Directional Flows of Capital 10
2.2 Signal of Credibility 12
3. Data 13
4. Methodology 17
5. Results 20
5.1 BITs and Foreign Capital Flows 22
5.2 BITs and Directional Flows of Capital 24
5.3 BITs as a Signal of Credibility 27
6. Robustness Checks 29
6.1 Variable Classification – Signing Date 30
6.2 Variable Classification – Institutional Quality 31
6.3 Sample Selection – M&A Dataset 32
6.4 Estimation Technique – Arellano-Bond Generalized Method of Moments 33
7. Conclusion 36
References 39
4
1. Introduction
“A BIT deal with China is on the Trump administration’s agenda”, said US Treasury Secretary Steven Mnuchin in June 2017 (Talley, 2017). This bilateral investment treaty (BIT) would give US firms broader access to the Chinese market and vice versa, on equal terms. After an exhaustive negotiation process between Washington and Beijing, it seems that the world’s two largest economies are on the brink of signing an important investment deal. But what are BITs and who exactly benefits from them?
Bilateral investment treaties are international agreements establishing the terms and
conditions for private investment by citizens and firms of one country into another (“Bilateral Investment Treaty”, n.d.). These treaties are designed to facilitate, promote and protect foreign direct investment (FDI) between the two states (Desbordes & Vicard, 2009). Ex-ante, the treaties establish transparency about risk and therefore reduce the risk of investing in a country. Ex-post, BITs ensure that firms have certain rights, for example property rights, and protect them from expropriation (Egger & Pfaffermayr, 2004). When a country ratifies a BIT, it is obliged to treat investors of the other signatory country at least as good as it treats its domestic investors. Next to that, privileges granted to one foreign investor must be granted to all foreign investors with the same nationality (Neumayer & Spess, 2005).
By entering into a BIT, the two states, in most cases, submit to binding arbitration through the United Nations Commission on International Trade Law (UNCITRAL) and the
International Centre for Settlement of Investment Disputes (ICSID).1 The ICSID, an affiliate
of the World Bank, was established in 1966 and has adjudicated on 700 investor-state
settlement disputes since its founding (World Bank, 2018a). Even though this caseload seems relatively small, the binding nature of the ICSID arbitration is considered to have had a large impact on investor-state dispute negotiations (Tobin & Rose-Ackerman, 2005).
At this moment, there are approximately 3000 signed BITs (UNCTAD, 2018). Yet the
literature in economics is inconclusive whether BITs actually stimulate foreign capital flows, and it provides conflicting theoretical reasoning for the claimed connection. For example, Salacuse and Sullivan (2005) and Busse, Königer, and Nunnenkamp (2010) find a strong positive correlation between BITs and FDI. At the same time Hallward-Driemeier (2003), Yackee (2010), and Chilton (2015) find little to zero evidence supporting this view and question whether the alleged benefits of entering into a BIT outweigh the costs of negotiating, signing, and complying with the treaty.
1 Even though most cases typically get adjudicated upon by these two arbitrary institutions, there are a few other institutions
5 While most studies on foreign capital flows use FDI as the dependent variable, a recent study by Bhagwat et al. (2017) questions the validity of this methodology. According to the authors, a considerable amount of FDI is channeled through offshore financial centers, concealing the true source or destination of the investment. They add that FDI contains other transactions that do not reflect investment decision directly, such as retained earnings and inter-company loans.
Consequently, Bhagwat et al. (2017) use cross-border mergers and acquisitions (M&A) as the dependent variable instead. They observe that the number and dollar amount of cross-border M&A roughly double after an investment treaty is signed. The authors find this effect to be concentrated in target countries with medium levels of political risk, thus supporting the view that BITs are ineffective for countries that face high levels of political risk and unnecessary for countries with low risk.
A sole focus on M&A activity, however, also has its drawbacks. M&A, for example, excludes greenfield investments, which are investments made by a firm to build its establishment in a foreign country (Nocke & Yeaple, 2007). Dixit (2011) notes that the construction of a new or greenfield plant can have a different economic effect than investment through merger or acquisition. I therefore choose not to analyze the effect of BITs on merely FDI or cross-border M&A, but study the effect on both of them, using two large comparable panel datasets.
The reason for concentrating on BITs is that the way they are implemented provides a suitable empirical framework for studying domestic institutions and foreign capital flows. BITs represent a powerful shock to usually slowly changing institutions, thereby providing large variation between countries in the protection and enforcement of contracts and investor property rights (Bhagwat et al., 2017). As the timing of the treaty might be endogenous to economic conditions, the dyadic nature of the BITs allows me to control for these other determinants that encourage investment in a country.
The research design in this study consists of multivariate fixed-effects regressions in order to control for factors that vary at the country-pair and year level. A simple cross-sectional regression does not suffice here, as this kind of regression only observes differences between countries at the same point of time, regardless of differences in time. Previous research highlights the importance of time-invariant factors such as geographical distance and
cultural differences in foreign investment decisions (e.g., Ahern, Daminelli, & Fracassi, 2015; Erel, Liao, & Weisbach, 2012). A fixed-effects estimation removes the effect of these time-invariant characteristics so that the net effect of BIT ratification on foreign capital flows can be assessed.
6 A stated purpose of BITs is that the ratification of a BIT directs the capital flow from
advanced (“developed”) economies to emerging (“developing”) economies (Egger &
Pfaffermayr, 2004). I find, however, that in the period 2001-2012, the effect of BITs on FDI is mostly concentrated at country-pairs consisting of two emerging economies. This result supports the view that firms incorporated in emerging countries, there being subject to poor governance, have developed skills to secure their outward FDI to other emerging countries better (Dixit, 2011, 2012). Robustness checks by means of changes in variable classification and sample size provide further support for this result.
Subsequently, I analyze the effect of BITs on foreign capital flows, conditional on
institutional quality. In the existing literature, there is no consensus whether BITs act as complements or substitutes for institutional quality. Consistent with Neumayer and Spess (2005) I find limited empirical evidence that BITs have a positive effect on FDI inflow for capital receiving countries with low institutional quality, but this result is not robust to different classifications of institutional quality. The conflicting findings of
Hallward-Driemeier (2003) and Tobin and Rose-Ackerman (2005, 2011), who claim that BITs rather act as complements to institutional quality, could be attributed to the fact that these studies exclude observations of FDI between two developing economies from their samples.
Regarding the estimations that use the number and dollar amount of cross-border M&A as the dependent variables, I do not obtain statistically significant results for the period 2001-2012. While FDI inflow has been gradually increasing during this period, M&A activity has been extremely volatile, as a result of the merger wave between 2002 and 2006 and the economic crisis that started in 2007. The discordance in findings with Bhagwat et al. (2017), who observe a positive impact of BITs on cross-border M&A, is likely due to differences in the sample period. To assess this “sample period effect”, I run the same estimations using a 1996-2006 cross-border M&A dataset. Strikingly, the coefficients then become statistically
significant and almost identical to the FDI dataset, confirming the presumption that the economic crisis had an enormous impact on the previous estimations.
This study contributes to three important kinds of literature. First, whether and how BITs affect cross-border flows of capital is still an unresolved issue (e.g., Egger & Pfaffermayr, 2004; Busse et al., 2010; Dixit, 2012). So far, academics have shared the view that FDI originates in advanced economies. However, at a time when developing economies are emerging as large investors (UNCTAD, 2009), this view appears to be outdated. Sharp increases in FDI outflow from economies such as China, Brazil, and India solicit a renewed analysis. My results show that the effect of BITs on FDI in the period 2001-2012 is actually most pronounced for foreign capital flows between such emerging economies.
7 Second, this study contributes to the literature on institutional quality and capital flows. Institutional quality is considered to cause substantial friction to cross-border investment (Papaioannou, 2009). According to Dixit (2012, p.1), weak governance generally lowers the amount of foreign capital inflow by “creating or worsening problems of commitment". My findings indicate that BITs might function as substitutes for good institutional quality, thus being most useful for countries that have low institutional quality standards.
Finally, the study contributes to the literature on capital flows in general. The presence of an international agreement is just one of the factors that affect investment decisions. Other factors, such as market size, openness to trade and inflation play a major role.2 In this paper,
I therefore control for these traditional determinants of FDI and M&A. To the best of my knowledge, this study is the first to perform an empirical analysis on the effect of BITs on comparable dyadic datasets of both FDI and cross-border M&A.
The paper is structured as follows. The next section illustrates how the importance of BITs grew over the years and discusses the theoretical framework. The hypotheses follow from this analysis. The data and the methodology are discussed in sections 3 and 4. Section 5 shows and discusses the results, after which section 6 presents robustness checks that address possible endogeneity in the research design. The final section concludes.
2. Theoretical Framework
Before the use of BITs, there were hardly any mechanisms to make state promises about the treatment of foreign investors credible (Elkins, Guzman, & Simmons, 2006). The traditional formulation of the customary international law on FDI, known as the “Hull Rule”, required “prompt, adequate and effective compensation” in the case of expropriation by the host government (Desbordes & Vicard, 2009, p.373). Besides the problem of enforcement, the rule did not allow countries to voluntarily signal their credibility (Elkins et al., 2006).
During the 1950s and 1960s, the Hull Rule was often challenged by Latin American countries and former colonies, and consensus on customary rules began to erode (“The Hull Rule”, n.d.).3 At this time, the first bilateral investment treaties were signed, with Germany and
Pakistan setting the bar in 1959 (UNCTAD, 2018). After that, it took almost twenty years before BITs gained momentum (Busse et al., 2010). While there were less than 400 BITs by the end of the 1980s, the number of BITs would grow to over 2,500 BITs in 2006 (Desbordes
2 For an overview see Chakrabarti (2001), Elkins, Guzman, and Simmons (2006), and Erel, Liao, and Weisbach (2012).
8 & Vicard, 2009). Figure 1 shows the number of BITs that were signed by region in the period 1980-2009.
Figure 1. The number of bilateral investment treaties signed per region. Source: Desbordes and Vicard (2009)
Contrary to the Hull Rule, BITs are more clearly focused on foreign investment protection. Firstly, by entering into a BIT, signatories agree to grant national treatment and most-favored nation treatment to foreign investors (Salacuse & Sullivan, 2005). National treatment means that foreign investors may not be treated any worse than national investors, but may be treated better. Most-favored nation treatment comprises that when privileges are granted to one foreign investor, these must be granted to all foreign investors.
Secondly, foreign investors are guaranteed “fair and equitable” treatment in accordance with international standards after the investment has taken place (ibid.). The countries also agree to guarantee free transfer and repatriation of capital and profits, compensation for
expropriated property or funds and protection of contractual rights. Finally, if an investment dispute arises, BITs generally make sure that the dispute gets adjudicated on by an
international arbitration body and not in the domestic country's court system (Bhagwat et al., 2017).
These dispute settlement provisions are perhaps the most crucial aspect of the BITs. Instead of attempting to litigate in the host country (of which the foreign investors may not like the quality and speed) or seeking diplomatic protection, investors now have recourse to an
9 international arbitration body, such as the ICSID and UNCITRAL. During arbitration
proceedings, three arbiters are selected, with each party selecting one and the institution selecting the third (Hallward-Driemeier, 2003). The decisions are often final; there is only a minimal possibility to appeal the decisions made by the court. Moreover, they cannot be amended by the domestic legal system or supreme court.
Despite this powerful legal framework, the ICSID has only been involved in about 700 cases since its founding (World Bank, 2018a). However, violation of a treaty will result in the reluctance of potential other signatory countries to sign further treaties and create
skepticism about the investment environment in the violating country. Thus, although the caseload at the ICSID has been relatively small, the chances are big that many disputes have been negotiated between parties because of the binding nature of ICSID arbitration (Tobin & Rose-Ackerman, 2005).
Of course, not all BITs contain identical provisions. Some countries, like the United States, insist on some limited rights for its investors for greenfield investments in the other
signatory country, whereas most BITs do not contain clauses for fair and equitable treatment while entering a new market.4 Conversely, some countries with emerging economies, such as
China, managed to restrict the compulsory dispute settlement provisions to only disputes regarding expropriation by the government (Bhagwat et al., 2017).
In general, however, BITs tend to be quite similar in their contents. Every treaty starts with both signatories declaring that the purpose of the treaty is to intensify the economic co-operation. The treaty between the Netherlands and Cambodia, ratified in 2006, for example reads: “… Desiring to strengthen their traditional ties of friendship and to extend and intensify the economic relations between them, particularly with respect to investments by the nationals of one Contracting Party in the territory of the other Contracting Party” (UNCTAD, 2005). After that, the different guarantees to and rights of investors are
discussed, such as national treatment and the dispute resolution mechanism. Furthermore, most BITs contain clauses that exclude investments in politically sensitive areas such as national security and the financial sector (Tobin & Rose-Ackerman, 2005).
It should be noted that the agreements in a BIT are reciprocal (Hallward-Driemeier, 2003). The rights given to investors from country A who invest in country B are the same as the rights given to investors of country B investing in country A. Does the ratification of a BIT then equally encourage foreign capital flows from country A to B and the other way around?
4 For instance, the US-Argentina BIT (Article XIV) reads, “This Treaty … shall apply to investments existing at the time of
10 In the next parts of this section, it is argued that the effect of a BIT on foreign capital flows is not equally distributed, as it is conditional on the direction of the investment and the level of institutional quality in the capital receiving country.
2.1
Directional Flows of Capital
The neoclassical growth theory, which assumes that countries have the same constant returns to scale production function, open world capital markets, and homogeneous capital and labor, predicts that capital should flow from rich to emerging countries, whereas the marginal product of capital is significantly higher in developing countries (Lucas, 1990). In reality, developed countries have always dominated the other countries in being recipients of foreign investments. This puzzle has become known as the Lucas Paradox and has generated a substantial body of literature.5Even though developed countries are still both the
dominating origin and the primary destination of FDI, this situation has been gradually changing, with developing countries receiving 36% of FDI in 2010 compared to only 20% in the 1980s (UNCTAD, 2004; Dixit, 2012). Interestingly, it was during this same period that the importance of BITs grew.
There are a few studies on BITs that study the effect of BITs on the directional flow of
capital. Using a non-dyadic dataset, Neumayer and Spess (2005) find that a higher number of BITs increases the amount of FDI that flows from developed to developing countries.
Notably, the authors exclude BITs signed between developing countries in their analysis “since FDI flows between developing countries are rare” (ibid., p. 17). Busse et al. (2010) use a gravity-type model on a bilateral FDI flow dataset and also conclude that BITs promote FDI flows to developing countries. Similar to Neumayer and Spess (2005), Busse et al. (2010) consider developing countries to be only capital receiving countries and not capital sources. This traditional view, though, which holds that the firm investing is based in a developed country and the recipient is a firm in a developing country, seems rather outdated. According to Dixit (2012), multinationals in developing economies have emerged as substantial
investors in other developing economies, and have begun to make investments in developed economies also.6
I wish to explore the possibility that the effect of BITs on foreign capital flows has shifted. That is, BITs are not effective anymore for advanced economies investing in emerging economies and have become most successful for emerging economies investing in other
5 Examples are Alfaro, Kalemli-Ozcan, and Volosovych (2008), Portes and Rey (2005), and Reinhart and Rogoff (2004). 6 Dixit refers to the acquisition of Arcelor by Mittal and Jaguar by Tata Motors as examples of firms in developing countries
11 emerging economies. This is in accordance with the idea outlined by Dixit (2011; 2012), who proposes that the experience of operating in challenging economic environments, and under poor governance institutions, helps firms in developing economies to cope with similar conditions in other developing economies. There are several explanations for this idea:
1. Firms in developing economies have adapted their production to the conditions of their country. Therefore, they are better in managing complex supply chains, unreliable power supplies, low-skilled workers, and so on.
2. These firms know how to overcome regulatory obstacles and have experience in dealing with weak contract enforcement.
3. The firms in developing countries are used to bribery in their own country, and more likely not to be constrained by domestic laws in their foreign operations.
4. Firms in developed countries face more (political) pressure in their countries to pay fair wages to their workers in developing countries.
Together, these explanations give firms from developing countries an advantage when investing in other developing economies. There are a few studies that empirically support this. Cuervo-Cazurra and Genc (2008) find that multinational firms from developing economies are more prevalent among the largest firms in countries that have more
corruption. Darby, Desbordes and Wooton (2010) find that worse governance at the capital receiving country leads to less FDI, but this negative effect is mitigated for FDI coming from developing countries. It should be noted that both of the aforementioned studies do not take the effect of ratification of a BIT into account.
However, regardless of the relative advantage of investors from developing economies, every investor needs some guarantee that he will not be expropriated by the government in the capital receiving country. It is widely recognized that economic globalization requires market-supporting institutions to flourish (Elkins et al., 2006). With respect to FDI, almost all the direct investments in foreign countries are governed by BITs. It is therefore expected that ratification of a BIT leads to higher foreign capital flows between the signatories and that this effect is most significant if the two signatories are developing economies. I hypothesize the following:
H1: Bilateral investment treaties increase foreign capital flows between two developing economies.
12
2.2
Signal of Credibility
Whilst several explanations for the Lucas Paradox exist, one considerable friction to cross-border investment is institutional quality (Papaioannou, 2009). There are various ways in which foreign capital flows can be affected by poor governance. For instance, foreign capital flows pertain to property rights for a stock of capital, not just a short-lived capital flow as is the case with a trade in goods (Dixit, 2012); the stakes are therefore a lot higher if
governance is weak. Another reason is that foreign capital flows often comprise large transactions that are non-repetitive and non-reversible, for example mergers and acquisitions. Hajzler (2012) and Colen, Persyn, and Guariso (2016) provide empirical
evidence that the risk of expropriation is increasing with the amount of sunk costs related to the investment.
Much empirical work exists on the connection between FDI and governance. For example, Globerman and Shapiro (2002) find that, after controlling for GDP and a couple of human development and environmental quality indices, better governance leads to more FDI inflows. Next to that, they show that developing countries tend to benefit more from
governance improvements than do richer countries. Using a firm-level dataset consisting of 22 transition economies, Javorcik and Wei (2009) find that corruption lowers FDI inflow and shifts the ownership structure towards joint ventures.
In addition, Anderson and Marcouiller (2002) and Leeson (2008) note that poor contract enforcement by a country’s legal system increases the cost of trade and thereby lowers the volume of trade. Thus, if a country wants to attract foreign investment, it has to signal that it is committed to protecting the investors. Ratification of a BIT represents a credible
commitment whereas there are considerable ex-post costs involved, such as diplomatic costs, sovereignty costs and arbitration costs (Elkins et al., 2006). When a country is subject to a BIT, it will be costly to treat the investors of the other signatory poorly (Tobin & Rose-Ackerman, 2005).
Academics, however, have found conflicting empirical evidence for whether BITs are complements or substitutes for good institutional quality. Neumayer and Spess (2005, p.5) find them to be substitutes as BITs “should provide security and certain standards of
treatment to foreign investors" in case domestic institutions do not offer these. Other studies (e.g., Tobin & Rose-Ackerman, 2011) argue that BITs might only be considered credible in an environment of good institutional quality. According to Hallward-Driemeier (2003), BITs have not provided a short-cut from the urge to invest in broader reforms of domestic institutions.
13 The problem with the studies mentioned in the previous paragraph is that they all assume that FDI originates in developed countries. Therefore, they do not include FDI flows between two developing economies. However, at a time when multinationals from developing
economies such as China, India, and Brazil are increasingly investing in other countries, this assumption does not hold anymore.
Cross-border investments add an extra element of insecurity that is not present in domestic investments (Dixit, 2011). The chances are higher that property rights are violated and that contracts are broken when there is poor institutional quality. Governments are more likely to violate such contracts with foreign investors as they are less worried about facing political consequences than if the investors were nationals (Bhagwat et al., 2017). On top of that, courts may be even biased in the case of disputes (Bhattacharya, Galpin, & Haslem, 2007). Foreign firms are less likely to invest in countries where they face political risk and
insecurity. The presence of a BIT, which aims at protecting foreign investors from this risk, can bring about investments that otherwise might not have been made. From the above theory I derive my second hypothesis:
H2: Bilateral investment treaties increase foreign capital flows to countries with weak domestic institutions since they function as substitutes for high institutional quality.
3. Data
This section describes the composition of the panel datasets that are used to test the hypotheses. Section 4 discusses the empirical specification.
Dependent Variables – FDI Inflow
The data on bilateral FDI flows is collected from the website of the United Nations Conference on Trade and Development (UNCTAD). The UNCTAD database uses the broadest measure of FDI, which means that it is composed of equity capital, reinvested earnings and intra-company loans (Tobin & Rose-Ackerman, 2005). The database covers the period 2001-2012, which is therefore the full sample period. All countries that do not have reported FDI data in the database are excluded. Next to that, the following two adjustments are made to the data:
14 1. Serbia and Montenegro were united as one country until 2006 under the names
“Yugoslavia” and “Serbia and Montenegro”.7 For this reason, there is no FDI data
available for the countries Serbia and Montenegro apart in the first years of the sample period. The countries Serbia, Montenegro, Yugoslavia, and Serbia and Montenegro are excluded from the sample to ensure data accuracy.
2. In order to reduce the skewness of the distribution of FDI inflow and thereby increase the fit of the model, it is common to take the natural log of this variable (e.g., Colen et al., 2016). However, as there are negative FDI flows in the data, the data first needs to be recoded (UNCTAD, 2001).8 One way to do this is to set all negative FDI flows
equal to positive $1, a method used by Neumayer and Spess (2005). This
transformation, however, may cause biases to the data; I therefore follow Tobin and Rose-Ackerman (2011) and use a log-transformation:
Log FDI = log(1+|FDI|), if FDI ≥ 0 Log FDI = − log(1+|FDI|), if FDI < 0
In the end, the sample consists of dyadic FDI data for 2362 country-pairs. I name the country from which capital flows to another country “capital source” and the country that receives the capital “capital recipient”. Note that I allow for separate observations for the country as a capital source and a capital recipient. That is, (Germany, Croatia, 2003) and (Croatia, Germany, 2003) are considered separate observations to allow for different country-level characteristics for the capital source and recipient.
Dependent Variables – Mergers and Acquisitions
The data on M&A is retrieved from Thomson One. I start with the entire database and set the following prerequisites:
1. The deal must have been completed in the period of January 1, 2001, to December 31, 2012.
2. The deal value must be publicly known.
3. The percentage of the stake in the firm after the transaction must be at least 20%, as this is considered the threshold for a shareholder to have a controlling interest.
7 Until 2003, this was the Federal Republic of Yugoslavia. Between 2003 and 2006, the countries formed the State Union of
Serbia and Montenegro.
15 Approximately 42,000 cross-border deals meet these conditions. Consistent with the FDI dataset, the countries Serbia, Montenegro, Yugoslavia, and Serbia and Montenegro are excluded from the dataset. Lastly, the country-pairs that are not present in the FDI dataset are excluded to ensure an equal amount of observations for both FDI and M&A.
Bilateral Investment Treaties
The data on bilateral investment treaties is retrieved from the online UNCTAD International Investment Agreements database (UNCTAD, 2018). The treaties that only contain
investment provisions (TIPs) are excluded because their contents may differ significantly. It should be noted that only country-pairs that did not have a ratified treaty yet in the year 2000 are included in the sample, to capture the so-called "treatment effect" of BIT ratification on foreign capital flows. It is unlikely that foreign capital flows in a country will increase immediately after signing of a BIT. Therefore, using the ratification date of a BIT is preferred over the signing date in the estimations. In some cases, an existing treaty was replaced by a new treaty. For these observations, the ratification date of the initial treaty is used as the relevant ratification date.
While almost every country signed their treaties by itself, Belgium and Luxembourg signed all their bilateral investment treaties through the Belgium-Luxembourg Economic Union (BLEU). In my sample, I assign all the treaties that the BLEU signed to both Belgium and Luxembourg separately.
Control Variables
The control variables are similar to the ones used in previous research (e.g., Tobin & Rose-Ackerman, 2011; Bhagwat et al., 2017) and control for factors that vary at the country-pair and year level. From the World Development Indicators database (World Bank, 2018b), I use the natural log of one plus gross domestic product (GDP) per capita, population, and
openness to trade as proxies for market size and market potential. Openness to trade is defined as the natural log of one plus the sum of imports and exports divided by a country’s GDP in a particular year.
As an indicator for macroeconomic stability, I use the difference between changes in
exchange rate against the US dollar for two countries (labeled as “Inflation rate”) as a control variable. That is, if the Argentinean peso in 2002 inflated 10% against the US dollar and the Indian rupee deflated 10% against the US dollar in the same year, the difference (1.1 minus 0.9) between the countries is included as a control variable. The data on inflation rates is retrieved from the Penn World Tables (Feenstra, Inklaar, & Timmer, 2015).
16 While the presence of natural resources in a country is expected to attract foreign investment regardless of otherwise discouraging factors, I also include a proxy for natural resource endowments, similar to the one used by Tobin and Rose-Ackerman (2005). This proxy is defined as the natural log of one plus the sum of fuel exports and ores and metals exports divided by the country’s GDP. More details on the construction and definition of each variable can be found in the appendix.
Worldwide Governance Indicators
Lastly, the Worldwide Governance Indicators from the World Bank are added to the model (World Bank, 2018c). The database comprises the following six indicators:
1. Voice and accountability: a variable that captures the extent to which a country's citizens can participate in selecting their government.
2. Rule of law: an indicator that captures perceptions of the extent to which agents have confidence in the quality of contract enforcement, property rights, the police, and the courts.
3. Regulatory quality: which measures the ability of the government to implement policies and regulations that permit and promote private sector development. 4. Political stability: a variable that captures perceptions of the likelihood of political
instability or politically-motivated violence.
5. Government effectiveness: which is an indicator of the quality of public services, civil service, and the degree of its independence from political pressures,
6. Control of corruption: a variable that captures perceptions of the extent to which public power is exercised for private gain.
Per indicator, the World Bank estimates a country's score on the aggregate indicator in units of a standard normal distribution, i.e., ranging from approximately -2.5 to 2.5. In order to obtain solely positive values, the scores are adjusted by adding 2.5 to them, which leads to an index of 0 to 5 per indicator. A score close to 0 means that a country is doing far worse than the average with respect to that indicator. Conversely, a score close to 5 means that a country outperforms most other countries.
17
4. Methodology
Before the hypotheses can be tested, an appropriate research design needs to be selected. Intuitively, a simple cross-sectional regression (“Pooled OLS”) does not capture the effect that I want to measure, as Pooled OLS only observes differences between countries at the same point of time, regardless of differences in time. In other words, a Pooled OLS analysis takes random samples in different time periods, and these samples will be populated by different countries. For example, it compares FDI data between the countries Argentina and India in 2002 with FDI data between Germany and Croatia in 2006.
The fundamental difference between Pooled OLS analyses and fixed/random-effects analyses is that in the latter case, the same countries get followed over time. That is, FDI data
between Argentina and India in 2002 is compared with FDI data between Argentina and India in 2006. I perform the Breush-Pagan Lagrangian multiplier test to determine
empirically whether a random-effects panel data model is preferred over Pooled OLS and find indeed that Pooled OLS is not an option for the analyses.9
Now that is confirmed that Pooled OLS cannot be used, I check whether there is a correlation between unobserved effects and the explanatory variables. I perform the Hausman test to see if the data is better suited for a fixed-effects or random-effects model.10As this test gives a
very low p-value, the null hypothesis of no correlation is rejected and fixed-effects are preferred over random-effects. The results of both tests can be found in the appendix.11
Several specifications are estimated to test the hypotheses. My base fixed-effects regression model for estimating the effect of BITs on foreign capital flows mirrors that of past studies:
Ln(1+FDI
ijt) = ∝
+ 𝛽
1Ln(1+FDI
ij(t-1)) + 𝛽
2Post-BIT
ij(t-1)+
𝛽
3(GDP/Capita)
it+
𝛽
4(GDP/Capita)
jt+ 𝛽
5Openness
it+ 𝛽
6Openness
jt+ 𝛽
7Inflation
(i-j)t+ 𝛽
8Population
it+
𝛽
9Population
jt+ 𝛽
10Resources
it+ 𝛽
11Resources
jt+∂
jt+ µ
ij+ 𝜏
t+ 𝜀
ij𝑡(1)
where the natural log of one plus FDI flows from capital source i to capital recipient j in time period t depends on the previous period’s FDI flow, a dummy indicating whether countries i and j ratified a bilateral investment prior to year t (Post-BIT), GDP per capita (GDP/Capita),
9 The “Breush-Pagan Lagrangian multiplier test” is used to test for heteroskedasticity in a linear model. The null hypothesis
states that there is no significant difference in variances of the error term between entities.
10 The Hausman test, also known as “Durbin-Wu-Hausman test”, tests whether the error term is correlated with the regressors.
The null hypothesis states that there is no correlation between the two.
11 Even though the results are not reported, the same tests were performed for the models with the number and dollar amount
18 the level of trade openness (Openness), the changes in inflation against the US dollar
(Inflation), the population (Population), the natural resource trade levels (Resources), the six different Worldwide Governance Indicators (∂jt
), country-pair fixed-effects (µ
ij), year fixed-effects (𝜏 t) and a random error term (𝜀 ij𝑡 ).It is preferred to use the natural log rather than the absolute level of FDI as the dependent variable, whereas the log specification reduces the weight of observations with very high FDI. On the right side of the equation, the one-period lagged value of FDI is included to account for sluggish adjustment of FDI, thereby following the approach of Colen et al. (2016). As regards the Worldwide Governance Indicators, only the indices of the capital recipient
country are included, as the index scores of the capital source country are presumed to have a negligible impact on the investment decisions of a foreign investor. Note that the models that take the dollar amount and the number of M&A as dependent variable look practically the same, except that the lagged FDI variable is replaced by either the lagged dollar amount or number variable.
Since country-pair and year fixed-effects are included in the model, the estimated effect can be defined as a within country-pair change over time. In case there is an omitted variable that (partly) explains my results, this cannot just be a country-specific or a country-year specific factor. The omitted variable must be a country-pair specific factor and vary over time to possibly lead to biased estimation results (Bhagwat et al., 2017). For example, economic changes in Viet Nam prior to the ratification of their BIT with Finland in 2009 would in itself not be able to explain an increase in capital flowing from Finland to Viet Nam in 2010, since these economic changes would also be present for a Norwegian company looking to invest in Viet Nam in 2010, with the only difference that Norway and Viet Nam did not have a ratified treaty in that year.
Even though model (1) is useful for understanding the independent effect of BITs on foreign capital flows, my theory proposes that this effect is conditional. In order to test whether bilateral investment treaties increase the amount of capital flowing from a developed economy to a developing economy, the country’s economies first need to be classified. I use the classification of the International Monetary Fund (IMF) to determine whether a country has a developed/advanced (labeled “Adv”) or a developing/emerging economy (“Emer”) (Nielsen, 2011).
The next step is to create dummy variables for each country-pair according to the direction of the capital flow. These capital flows can be classified as either “Advanced-Advanced”,
19 FDI flow from the United Kingdom, which has an advanced economy, to emerging economy Nigeria is defined as “Adv-Emer”. I do not allow for changes of a country switching from an emerging economy to an advanced economy in the sample period. Thus, whether an economy has an emerging or advanced economy does not depend on time. Lastly, the directional flow indicator is interacted with the Post-BIT dummy. The model for testing the first hypothesis is shown below:
Ln(1+FDI
ijt) = ∝
+ 𝛽
1Ln(1+FDI
ij(t-1)) + 𝛽
2Post-BIT
ij(t-1)+
𝛽
3Post-BIT
ij(t-1)* Adv-Emer
ij+
𝛽
4Post-BIT
ij(t-1)* Emer-Adv
ij+ 𝛽
5Post-BIT
ij(t-1)* Adv-Adv
ij+ 𝛽
6(GDP/Capita)
it+
𝛽
7(GDP/Capita)
jt+ 𝛽
8Openness
it+ 𝛽
9Openness
jt+ 𝛽
10Inflation
(i-j)t+ 𝛽
11Population
it+
𝛽
12Population
jt+ 𝛽
13Resources
it+ 𝛽
14Resources
jt+∂
jt+ µ
ij+ 𝜏
t+ 𝜀
ij𝑡(2)
where the coefficients 𝛽2 to 𝛽5 represent the effect of BITs on FDI, conditional on the directionof the FDI flow. The interaction variable with the “Emerging-Emerging” dummy is left out of the model to prevent perfect multicollinearity. Coefficient 𝛽2 thus demonstrates the effect of
BITs on capital flowing between two emerging economies.
In order to test the second hypothesis, I use a similar model to model (2), except that the interaction variables are replaced. This time, I compute a proxy for institutional quality using the World Bank’s Worldwide Governance Indicators. As was mentioned in section 3, the six indicators are slightly adjusted which results in index scores ranging from 0 to 5. Based on a total score of 0 to 30, the capital receiving countries are classified into two classes: Low institutional quality (0-16.3); High institutional quality (16.3-30). A score of 16.3 is chosen as the threshold as this is the average institutional quality score in the sample.
Subsequently, the institutional quality indicator is interacted with the Post-BIT dummy. The interactive model then looks as follows:
Ln(1+FDI
ijt) = ∝
+ 𝛽
1Ln(1+FDI
ij(t-1)) + 𝛽
2Post-BIT
ij(t-1)+
𝛽
3Post-BIT
ij(t-1)* High Inst.
Quality
ij+ 𝛽
4(GDP/Capita)
it+ 𝛽
5(GDP/Capita)
jt+ 𝛽
6Openness
it+ 𝛽
7Openness
jt+
𝛽
8Inflation
(i-j)t+ 𝛽
9Population
it+ 𝛽
10Population
jt+ 𝛽
11Resources
it+ 𝛽
12Resources
jt+∂
jt+
µ
ij+ 𝜏
t+ 𝜀
ij𝑡(3)
where coefficient 𝛽 2 and 𝛽 3 represents the effect of BITs on FDI, conditional on the
institutional quality of the capital recipient. Similar to model (2), the interaction variable with the “Low institutional quality dummy” is left out of the model to prevent perfect multicollinearity with the “High institutional quality dummy”.
20
5. Results
Before discussing the multivariate models that were presented in the previous section, I first analyze the average univariate effect of ratifying a bilateral investment treaty on FDI flows and M&A activity. Table 1 shows the results of univariate t-tests that assess this effect. The unit of observation is a capital source i, capital recipient j, and year t. Column (C) shows the p-value for the difference in the mean before and after the ratification of a BIT.
About 16% of the observations in the sample are after the ratification year of a BIT. It can be observed that the log FDI inflow increases from 1.058 to 1.247, and this difference is
significant at the 5% level. From an absolute perspective, the mean of FDI inflow increases from €44 million to €72 million, which is a large economic effect. Surprisingly, I find reversed results for the other two dependent variables, i.e., the dollar amount and the number of M&A. The table shows that there is a significant drop in M&A activity after the ratification of a BIT. The means of both the dollar amount and number of M&A are considerably lower for Post-BIT observations, and the differences in the means are statistically significant at the 1% level. The increase in FDI versus the decrease in M&A activity suggests that these
measures of capital flows cannot be compared one-to-one.
A possible explanation is that M&A activity was very much affected by the economic crisis, which started exactly in the middle of the sample period. Moreover, the crisis period was preceded by a merger wave (Lipton, 2006). As most Post-BIT observations are at the end of the sample period, thus after the start of the economic crisis, it is not surprising that the means of the number and dollar amount of M&A decreased for Post-BIT observations. In section 6 different time periods of the M&A dataset are analyzed in more detail.
Table 1 reports diversified results for the economic control variables. For the capital source, there is only a significant difference in the mean of the population variable. Meanwhile, the differences in the means of openness and natural resource endowments are significant for the capital recipient. There seems to be no difference for the inflation rate variable. At this stage, it is difficult to draw conclusions based on these coefficients. Something that does stand out in the univariate analysis is that the means of the Worldwide Governance Indicators of the capital recipient are all significantly lower after the ratification of a BIT than before. This is in line with the expectation that foreign investors are more willing to invest in countries that are considered to have lower institutional quality after a bilateral investment treaty is in place. Section 5.3 elaborates on institutional quality.
21 It must be stressed that in the univariate analysis only one variable at a time is analyzed; there are no controls for time trends or country-pair differences. These differences are controlled for in the subsequent analyses.
Table 1. Summary statistics
This table reports summary statistics for the years before and after the ratification of a bilateral investment treaty. Each observation is a capital source i, capital recipient j, and year t combination. The mean and standard deviation (in parentheses) for each variable are reported separately in columns A and B. Column C reports the p-value from a t-test of the difference between Pre-BIT and Post-BIT years, with the standard errors adjusted for clustering at the country-pair level (*** p<0.01, ** p<0.05, * p<0.1). The lagged values of FDI and M&A activity (“deals”) and the Post-BIT indicator are not reported. Please see the appendix for definitions of all variables.
(A) (B) (C)
Pre-BIT Years Post-BIT Years p-value of difference
Ln(1+FDI Inflowijt) 1.058 1.247 0.044**
(0.035) (0.087) Ln(1+$dealsijt) 0.601 0.230 0.000*** (0.030) (0.074) Ln(1+#dealsijt) 0.141 0.055 0.000*** (0.007) (0.018) Ln(1+(GDP/Capita))it 9.530 9.511 0.813 (0.030) (0.076) Ln(1+(GDP/Capita))jt 8.728 8.762 0.696 (0.033) (0.083) Ln(1+(Openness))it 0.635 0.662 0.138 (0.007) (0.017) Ln(1+(Openness))jt 0.562 0.587 0.038** (0.004) (0.011) Inflation(i-j)t -0.010 -0.005 0.176 (0.001) (0.003) Ln(1+(Population))it 2.809 3.089 0.007*** (0.038) (0.096) Ln(1+(Population))jt 2.932 3.056 0.174 (0.034) (0.085) Ln(1+(Resources/GDP))it 0.064 0.073 0.133 (0.002) (0.005) Ln(1+(Resources/GDP))jt 0.057 0.067 0.046** (0.002) (0.005)
Voice and Accountabilityjt 2.810 2.646 0.003***
(0.020) (0.050) Rule of Lawjt 2.712 2.573 0.017** (0.022) (0.056) Regulatory Qualityjt 2.867 2.745 0.025** (0.020) (0.050) Political Stabilityjt 2.501 2.402 0.053* (0.020) (0.051) Government Effectivenessjt 2.834 2.698 0.017** (0.021) (0.054) Control of Corruptionjt 2.726 2.489 0.000*** (0.023) (0.058) Observations 24443 3901
22
5.1
BITs and Foreign Capital Flows
Table 2 shows the results of the first multivariate model, in which I include economic and geographic control variables that vary at the country-pair and year level such as GDP per capita, openness to trade, population and so on. Column (A) shows the estimation results with FDI as the dependent variable; columns (B) and (C) demonstrate model (1) using the dollar amount and the number of M&A (“deals”) as dependent variables.
Column (A) in table 2 shows that for the estimation with FDI inflow as the dependent
variable, the Post-BIT coefficient is positive and significant at the 5% level, indicating a 0.151 log point increase in FDI inflow after the ratification of a bilateral investment treaty. This increase is an economically large effect, given that the mean log of FDI inflow before treaty ratification is 1.058.
Interestingly, I do not observe a similar effect for the estimations in columns (B) and (C). On the contrary, the coefficients of the Post-BIT indicator for the M&A models are even slightly negative, respectively -0.014 log point for the dollar amount and -0.003 log point for the number of deals. Both coefficients, however, are statistically insignificant and relatively small compared to the mean logs that were shown in the summary statistics.
With regard to the control variables, it can be observed that capital is clearly flowing to larger markets, whereas the coefficients for the market size proxies (GDP per capita and population) are positive and in most cases statistically significant. The coefficients of the inflation variable in columns (A), (B), and (C) all have a negative sign; this was to be expected, as higher inflation rates can be associated with lower capital flows (Busse et al., 2010). It should be noted, however, that these coefficients do not reach the conventional significance level. The coefficients of the other control variables are ambiguous they have alternate positive and negative signs.
In table 2, no distinction is being made between countries with an advanced or emerging economy or between countries with different kinds of institutional quality. In order to confirm or reject the hypotheses, I have to look into the effect of BITs on capital flows, conditional on the type of economy and the degree of institutional quality. The next two sections cover the conditionality of these items.
23
Table 2. Effect of BITs on foreign capital flows
This table reports results from fixed-effects regressions (model 1) where the unit of observation is a capital source i, capital recipient j, and year t combination. Column A shows the estimation results using the natural log of one plus the FDI inflow as the dependent variable. In column B and C, the natural logs of one plus the dollar amount and number of deals are used as dependent variables. Post-BIT is an indicator equal to 1 for all years after the ratification of a Post-BIT between countries i and j. All three models include country-pair and year fixed-effects. The constants are not reported. The standard errors are clustered at the country-pair level and are reported in parentheses (*** p<0.01, ** p<0.05, * p<0.1).
(A) (B) (C)
Ln(1+FDI Inflowijt) Ln(1+$dealsijt) Ln(1+#dealsijt)
Ln(1+FDI Inflowij(t-1)) -0.009
(0.014) Ln(1+$dealsij(t-1)) -0.002 (0.016) Ln(1+#dealsij(t-1)) 0.014 (0.018) Post-BIT 0.151** -0.014 -0.003 (0.077) (0.036) (0.008) Ln(1+(GDP/Capita))it 0.853*** 0.093 0.022* (0.137) (0.057) (0.012) Ln(1+(GDP/Capita))jt 0.794*** 0.120** 0.031*** (0.138) (0.056) (0.012) Ln(1+(Openness))it 0.275 -0.322** -0.053 (0.436) (0.158) (0.035) Ln(1+(Openness))jt 1.129*** -0.076 0.007 (0.436) (0.149) (0.033) Inflation(i-j)t -0.024 -0.043 -0.004 (0.109) (0.042) (0.008) Ln(1+(Population))it 0.943** 0.390** 0.071* (0.471) (0.174) (0.038) Ln(1+(Population))jt 4.279*** 0.151 0.055 (0.616) (0.192) (0.039) Ln(1+(Resources/GDP))it 1.479* 0.315 0.001 (0.785) (0.250) (0.053) Ln(1+(Resources/GDP))jt -1.701** 0.811*** -0.128** (0.845) (0.284) (0.059)
Voice and Accountabilityjt -0.270* -0.030 -0.012
(0.144) (0.047) (0.010) Rule of Lawjt -0.359* 0.041 -0.001 (0.183) (0.064) (0.013) Regulatory Qualityjt -0.083 0.065 0.021* (0.166) (0.056) (0.012) Political Stabilityjt 0.206*** 0.028 0.007 (0.079) (0.029) (0.006) Government Effectivenessjt 0.711*** 0.004 -0.005 (0.198) (0.077) (0.016) Control of Corruptionjt -0.108 0.046 0.020 (0.165) (0.060) (0.013) Country-pair FE Year FE Observations R-squared Yes Yes 22,066 0.320 Yes Yes 22,066 0.712 Yes Yes 22,066 0.769
24
5.2
BITs and Directional Flows of Capital
While the prior analysis demonstrates that ratification of a BIT can be associated with higher FDI inflows, it does not look into the mechanisms that lead to this increase. As mentioned before, the traditional view on BITs holds that they should encourage foreign capital flows from economies with abundant capital and skilled labor to less developed economies (Egger & Pfaffermayr, 2004).
However, it seems that this view of a multinational firm in a developed economy investing in developing economies is getting more and more obsolete. Multinational firms in developing countries have emerged as substantial investors, and have even started investing in
developed countries (Dixit, 2012). In the World Investment Report 2009, it is pointed out that about 16% of the global FDI outflows in 2008 originated in a developing country (UNCTAD, 2009). Has the flow of foreign capital reversed direction?
Table 3 assesses the effect of BITs on foreign capital flows, conditional on the type of
economy. The Post-BIT dummy is interacted with the capital direction dummy (for example, "Emerging-Advanced" would equal 1 for a capital flow from Bolivia to the United States, and 0 otherwise) to test whether there are differential effects of BITs on the direction of foreign capital flows.
Consistent with hypothesis 1, column (A) demonstrates that almost the entire increase in FDI inflow after the ratification of a BIT is concentrated at capital flows between two emerging economies. The Post-BIT coefficient, which represents the “Emerging-Emerging” flows, is significant at the 1% level and with 0.478 log point a lot bigger than the
unconditional effect of BITs on FDI (resp. 0.151 log point). The coefficients of the interaction dummies “Advanced-Emerging” and “Emerging-Advanced” are both negative and significant at the 5% level. Note that these coefficients should be interpreted using the Post-BIT
coefficient as starting point; thus the impact of BITs on "Advanced-Emerging" flows is with 0.478 - 0.374 = 0.104 log point positive, albeit not significantly different from zero.
The split between advanced and emerging economies does not lead to noticeable results for the M&A estimations in columns (B) and (C). Even though the coefficients for most of the directional flow indicators are reversed, they cannot properly be interpreted since all coefficients are statistically insignificant.
25
Table 3. Direction of foreign capital flows
This table reports results from fixed-effects regressions (model 2) where the unit of observation is a capital source i, capital recipient j, and year t combination. The dependent variables are indicated in the column title. I follow IMF’s classification and divide the sample countries into two groups: Advanced economies and Emerging economies. For example, “Emerging-Advanced” is a dummy
variable equal to 1 if the foreign capital flows from an emerging economy to an advanced economy. The other dummy variables are similarly defined. All three models include country-pair and year fixed-effects. The constants are not reported. The standard errors are clustered at the country-pair level and are reported in parentheses (*** p<0.01, ** p<0.05, * p<0.1).
(A) (B) (C)
Ln(1+FDI Inflowijt) Ln(1+$dealsijt) Ln(1+#dealsijt)
Ln(1+FDI Inflowij(t-1)) -0.009
(0.014) Ln(1+$dealsij(t-1)) 0.002 (0.016) Ln(1+#dealsij(t-1)) 0.014 (0.018) Post-BIT 0.478*** -0.056 -0.008 (0.150) (0.063) (0.015) Post-BIT X Advanced-Emerging -0.374** 0.063 0.008 (0.186) (0.080) (0.018) Post-BIT X Emerging-Advanced -0.417** 0.028 -0.001 (0.184) (0.083) (0.018) Post-BIT X Advanced-Advanced -0.670 0.118 0.040 (0.484) (0.211) (0.050) Ln(1+(GDP/Capita))it 0.834*** 0.099* 0.024* (0.138) (0.057) (0.012) Ln(1+(GDP/Capita))jt 0.781*** 0.119** 0.030** (0.139) (0.056) (0.012) Ln(1+(Openness))it 0.281 -0.323** -0.054 (0.436) (0.158) (0.035) Ln(1+(Openness))jt 1.115** -0.075 0.006 (0.440) (0.147) (0.032) Inflation(i-j)t -0.026 -0.043 -0.004 (0.109) (0.042) (0.008) Ln(1+(Population))it 0.860* 0.410** 0.075** (0.473) (0.175) (0.038) Ln(1+(Population))jt 4.161*** 0.154 0.055 (0.625) (0.195) (0.040) Ln(1+(Resources/GDP))it 1.489* 0.311 -0.001 (0.787) (0.250) (0.053) Ln(1+(Resources/GDP))jt -1.701** -0.816*** -0.128** (0.846) (0.284) (0.059)
Voice and Accountabilityjt -0.267* -0.030 -0.012
(0.144) (0.047) (0.010) Rule of Lawjt -0.340* 0.040 -0.001 (0.184) (0.064) (0.013) Regulatory Qualityjt -0.086 0.065 0.021* (0.166) (0.057) (0.012) Political Stabilityjt 0.205*** 0.028 0.007 (0.079) (0.029) (0.006) Government Effectivenessjt 0.707*** 0.003 -0.006 (0.198) (0.077) (0.016)
26 Control of Corruptionjt -0.112 0.047 0.020 (0.165) (0.060) (0.013) Country-pair FE Year FE N R-squared Yes Yes 22,066 0.320 Yes Yes 22,066 0.712 Yes Yes 22,066 0.769 The split between advanced and emerging economies does not lead to noticeable results for the M&A estimations in columns (B) and (C). Even though the coefficients for most of the directional flow indicators are reversed, they cannot properly be interpreted since all coefficients are statistically insignificant.
Overall, the results in table 3 provide reasonable empirical evidence to support hypothesis 1. While previous studies (e.g., Neumayer & Spess, 2005; Bhagwat et al., 2017) observe an increase in capital flows from developed to developing economies after BIT ratification, this effect is not observable in my dyadic FDI and M&A datasets. In fact, most capital seems to flow from emerging economies to other emerging economies, but this finding only holds for the estimation with the natural log of one plus FDI as the dependent variable.
The finding that the effect of BITs on FDI is concentrated at “Emerging-Emerging” flows goes against existing evidence. How should this be interpreted? As discussed in section 2.1, the proposition by Dixit (2012) comprises that multinational firms in developing countries have learned how to cope with difficult conditions and are able to use this in their advantage. Besides this competitive advantage, academics have found several other explanations for the increasing outward FDI flows from developing countries. Dunning (2008) for example notes that multinational firms from emerging markets tend to internationalize earlier than their developed counterparts due to the globalization process.
Another possible explanation is that firms in developing countries tend to be driven by strategic and political rather than economic motivations (Sauvant, McAllister, & Maschek (eds.), 2010). These firms, often encouraged or subsidized by their home governments, use FDI to acquire natural resources or land rights in other developing countries (Lall, Chen, Katz, Kosacoff, & Villela, 1983). The recent Chinese investments in Africa and the Middle East reflect this.
A final question that arises is why previous research finds the effect of BITs on FDI and M&A activity to be concentrated at “Advanced-Emerging” flows. This can most likely be attributed to differences in the sample composition and sample period. More specifically, Neumayer and Spess (2005, p.17) exclude BITs signed between developing countries “since FDI flows between developing countries are rare”. According to the authors, this adjustment is defensible as their study covers the period 1970-2001, a period in which there was not
27 much FDI between developing countries. Busse et al. (2010) analyze roughly the same period as Neumayer and Spess (2005) and also exclude FDI outflow from developing countries. The prevalence of FDI originating in developed economies in the 70s, 80s, and 90s probably causes the authors to find the effect of BITs on FDI to be most pronounced for “Advanced-Emerging” flows. Although Bhagwat et al. (2017) use a different sample period (1980-2012), the inclusion of the 80s and the 90s in the sample period presumably has a large impact on their findings as well. The extent to which the sample period affects the results is elaborated on in section 6.3.
5.3
BITs as a Signal of Credibility
It is often argued that BITs function as a mechanism to signal credible commitments. After all, if a country is subject to many BITs, it will be costly to treat foreign investors poorly (Tobin & Rose-Ackerman, 2005). In case a country has weak domestic institutions, it risks costly litigation if they choose to expropriate foreign investors. BITs should therefore provide countries with an incentive to invest in high-quality institutions. The next part of this section assesses this topic.
To differentiate between institutional quality, the capital receiving countries are classified into two groups: those with high institutional quality and those with low institutional quality. A country is considered to have high institutional quality if it has an institutional quality score between 16.3 and 30, based on the six Worldwide Governance Indicators. If the score is below 16.3, the country is classified as a low institutional quality country. The threshold of 16.3 is chosen because this is the average institutional quality score in the sample. Similar to model (2), the institutional quality indicator is interacted with the Post-BIT dummy to capture the effect of Post-BITs on foreign capital flows, conditional on the level of institutional quality. The interaction variable with the “Low institutional quality” dummy is left out of the estimation to prevent perfect multicollinearity. The coefficient "Post-BIT" represents the effect of this dummy.
From table 4, it can be observed that the coefficient of Post-BIT in column (A) (resp. 0.214) is positive and statistically significant at the 5% level. This result supports the second
hypothesis; ratification of a BIT has a positive effect on foreign capital flows towards
countries that have inferior standards of institutional quality. The effect is also economically large, given that the unconditional effect of BITs on FDI was only 0.151 log point. Similar to the other tables, the estimations in column (B) and (C) with M&A as the dependent variable do not provide statistically significant results.
28
Table 4. Foreign capital flows and institutional quality
This table reports results from fixed-effects regressions (model 3) where the unit of observation is a capital source i, capital recipient j, and year t combination. The dependent variables are indicated in the column title. High Inst. Quality is an indicator equal to 1 if the capital receiving country has an institutional quality score between 16.3 and 30, and equal to 0 otherwise. If the capital recipient has a score below 16.3, the country is considered to have a low institutional quality standard. All three models include pair and year fixed-effects. The standard errors are clustered at the country-pair level and are reported in parentheses (*** p<0.01, ** p<0.05, * p<0.1). The constants are not reported.
(A) (B) (C)
Ln(1+FDI Inflowijt) Ln(1+$dealsijt) Ln(1+#dealsijt)
Ln(1+FDI Inflowij(t-1)) -0.009
(0.014) Ln(1+$dealsij(t-1)) 0.002 (0.016) Ln(1+#dealsij(t-1)) 0.014 (0.018) Post-BIT 0.214** -0.005 -0.000 (0.099) (0.045) (0.010)
Post-BIT X High Inst. Quality -0.157 -0.022 -0.006
(0.142) (0.068) (0.015) Ln(1+(GDP/Capita))it 0.861*** 0.094* 0.023* (0.137) (0.057) (0.012) Ln(1+(GDP/Capita))jt 0.783*** 0.119** 0.030** (0.139) (0.056) (0.012) Ln(1+(Openness))it 0.282 -0.321** -0.053 (0.436) (0.157) (0.035) Ln(1+(Openness))jt 1.124** -0.077 0.007 (0.437) (0.149) (0.033) Inflation(i-j)t -0.025 -0.043 -0.004 (0.109) (0.041) (0.008) Ln(1+(Population))it 0.956** 0.392** 0.071* (0.470) (0.174) (0.038) Ln(1+(Population))jt 4.187*** 0.138 0.051 (0.627) (0.196) (0.040) Ln(1+(Resources/GDP))it 1.468* 0.314 0.000 (0.786) (0.250) (0.053) Ln(1+(Resources/GDP))jt -1.727** -0.815*** -0.129** (0.846) (0.284) (0.059)
Voice and Accountabilityjt -0.268* -0.029 -0.012
(0.144) (0.047) (0.010) Rule of Lawjt -0.345* 0.042 -0.000 (0.184) (0.064) (0.013) Regulatory Qualityjt -0.088 0.064 0.021* (0.166) (0.056) (0.012) Political Stabilityjt 0.206*** 0.028 0.007 (0.079) (0.029) (0.006) Government Effectivenessjt 0.704*** 0.003 -0.005 (0.199) (0.077) (0.016) Control of Corruptionjt -0.110 0.046 0.020 (0.165) (0.060) (0.013) Country-pair FE Year FE N R-squared Yes Yes 22,066 0.320 Yes Yes 22,066 0.712 Yes Yes 22,066 0.769
29 The findings in column (A) are as to be expected. Table 3 already showed that there is a positive correlation between the ratification of a BITs and FDI in emerging economies. Intuitively, institutional quality in these emerging economies is lower than in developed economies. The effect of BITs on FDI inflow in countries with poor institutional quality is thus expected to mirror the effect for emerging economies. Comparing the means of the institutional quality scores for advanced economies (22.3) and emerging economies (12.8) in the sample confirms this.
Overall, the findings in table 4 indicate that BITs are most effective for countries that have low institutional quality standards and that they are not necessary for countries with high institutional quality. This is in line with Neumayer and Spess (2005), who also provide limited empirical evidence that BITs act as substitutes for institutional quality, and
Ginsburg (2005), who argues that, under some circumstances, international devices such as BITs may substitute for local institutions.
My results, however, contradict the findings by Hallward-Driemeier (2003) and Tobin and Rose-Ackerman (2011), who claim that BITs are only credible if a country already has the necessary domestic institutions in place. A possible explanation for the difference in findings is that these two studiesexclude foreign capital flows between two developing economies. Their motivation is that in the 80s and 90s there was not much FDI outflow from developing countries. Tobin and Rose-Ackerman (2011, p. 5) already noted that the results of their paper would have to be revised from the moment that countries such as India, China, and Brazil would increase their FDI outflow. Another reason for the discrepancy in the results might be the way that the countries are classified. In the next section, different classifications of institutional quality are used to check if the results in table 4 are robust.
6. Robustness Checks
The combined estimation results of section 5.2 and 5.3 fit with many parts of the proposition that are stated by Dixit (2011 & 2012). The effect of BITs on FDI is no longer observed for capital flows from developed to developing countries, but is now in fact concentrated at FDI flows between two developing economies. Also, BITs seem to be most effective for countries with low institutional quality. To address potential endogeneity concerns related to the research design, this section performs robustness checks of changes in the classification of variables, sample size, and estimation technique.
