The Idiosyncratic Relationship Between Brazil and Idiosyncratic Relationship Between Brazil and BITs:…

  • Published on

  • View

  • Download


  • The Idiosyncratic RelationshipBetween Brazil and BITs: A synthetic

    control approachPaulo Cavallo

    University of Texas at Dallas

    January 17, 2017


    The explosion in Bilateral Investment Treaties (BITs) signed between countries in the 1990s and theconcurrent surge in FDI flows drew substantial attention in the literature. The subsequent studies ontheir relationship have found conflicting results on the impact of BITs on FDI inflows. This paper attemptsto tackle the controversial relationship between BITs and FDI inflows by adapting an innovative techniqueto analyze Brazils unique stance towards BITs: the synthetic control method. Brazil is a peculiar case inthe BIT system in that it defies common belief that entering into BITs is necessary for attracting FDI. Thispaper seeks to estimate the impact of Brazils exclusion from the BIT system in the level of FDI inflowinto the country between 1990 and 2010 by generating a synthetic control group that reproduces BrazilsFDI inflows after the surge in BITs in the early 1990s. There is evidence that, although Brazil did receivesubstantial amounts of FDI even in the absence of BITs, have they joined the rest of the world and enteredinto BITs with other countries the inflow in the period would have been greater. This holds despite thelarge privatizations in Brazil in the late 1990s that attracted substantial international attention and FDI.The findings are robust to both in-space and in-time placebo experiments.

    I. Introduction

    Over the last two decades, the surge in volume of FDI flows in the world was accompaniedby the eruption of Bilateral Investment Treaties (BITs) being signed between countries.This surge in BITs could be seen as a normal reaction to the higher volumes of FDI as thetreaties primary raison detre is to protect foreign investors. This surge can also be seen as thecause of higher FDI inflows to a given country due to the important signals these treaties send tointernational investors. As a result of these signals, nearly 2,500 Bilateral Investment Treaties hadbeen signed worldwide between 1991 and 2000 in an effort to protect foreign investors (UNCTAD2001). Not surprisingly, BITs became a highly contentious focus of the literature studying theirimpact on the flow of FDI in and out of countries with mixed results so far. This paper willattempt to compare the FDI inflows of a country that has been affected by BITs with the inflowsof a country that has not. For this to happen, firstly we would need a country that has not beenaffected in any way by the influence of BITs on FDI flows. The problem is BITs are widespreadthroughout the world and finding a comparison unit that has not entered into any BITs wouldbe difficult. More importantly, even if we find a country that has no BITs in place, finding acomparison unit would still be an impossible task as no country is similar to another.

    School of Economic, Political and Policy Sciences - UT Dallas


  • There are though a handful of countries that have not been affected by the wave of BITs thatswept through the world in the 1990s. One of these countries is a big player in the internationalinvestment system and has defied the BIT logic by attracting massive amounts of FDI withoutever having enacted a BIT; Brazil. The second task would be harder to fulfill as, since no countryis comparable to another in all the factors that would determine its level of FDI inflow, we wouldnot find a suitable comparison unit to Brazil. In light of this issue, Abadie et al. (2010) haveput forward a data-driven procedure to construct suitable comparison groups in such situations:the Synthetic Control Method for Comparative Case Studies1. The synthetic control builds aweighted average of the available control units while making explicit the relative contributionof each control unit to the synthetic unit and the similarities (or lack thereof) between them interms of preintervention outcomes and other predictors of postintervention outcomes (Abadieet al. 2010). In order to be able to apply the synthetic control we need one affected unit, severalcomparable unaffected units and a number of predictors to build the synthetic unit. We also needan event or a specific policy or intervention that happens at a single point in time and allows us tosee where the synthetic unit starts diverging from the original affected unit. For instance, Abadie& Gardeazabal (2003) investigate the economic effects of conflict using the terrorist conflict in theBasque Country as a case study2.

    During the 1990s in most of Latin America, we saw, parallel to the widespread process ofeconomic liberalization, a multiplication and rapid dissemination of BITs. Yet, regardless of thisgrowth in the importance of BITs and their possible connection with FDI, curiously enough, Brazilhas remained out of the BIT international system. Brazil is the only country in South America, andthe only major country in the world, that has absolutely no BIT in place. Brazil is often seen asthe exception to the ever so widespread rule that BITs do help countries attract more investments.Brazilian officials themselves argued at the time that Brazil did not need a tool that would infringeits sovereignty as it was already attracting massive volumes of FDI without them. Yet, is that reallythe case? Is Brazil really an exception to the rule? Was Brazil right in assuming it did not needBITs to attract FDI? And, more importantly, would Brazil have attracted even higher volumes ofFDI had it succumbed to the BITs fever taking hold of the world in the 1990?

    Brazils unique antagonistic position towards BITs, the substantial amount of FDI it receivedregardless of this position and the sudden surge in BITs in the early 1990 offer us an incredibleopportunity to apply the synthetic control method and contribute in a novel way to the literature.This paper will not introduce any new covariates to a specific model or propose they are measureddifferently. The comparison between the FDI inflows into the actual Brazil, with no BITs in force,and the synthetic Brazil with BITs will shed light into the impact of BITs on FDI inflows in thecountry. This analysis will allow us to investigate the impact of BITs on one of the very few caseswhere BITs and FDI do not go hand in hand therefore contributing in a different way to the studyof the impact of BITs on FDI flows. In doing so, the paper will also help solve a puzzle that hasintrigued the scholars of international investment for some time, the idiosyncratic relationshipbetween Brazil and BITs.

    The findings of the analysis carried out on this paper suggest Brazil could have attracted highervolumes of FDI between 1990 and 2010 had it enacted any treaty. These findings are robust toboth in-time and in-space placebo experiments as well as a number of robustness checks carriedout to increase confidence in the model and method used. These findings can also be seen asevidence of the correlation between BITs and FDI, even though they were obtained in a muchdifferent analytic way than the bulk of the literature has seen so far. This paper also contributes

    1See Abadie & Gardeazabal (2003), Abadie et al. (2010), Abadie et al. (2015)2Abadie & Gardeazabal (2003) use other Spanish regions to construct a synthetic control region which resembles

    relevant economic characteristics of the Basque Country before the outset of Basque political terrorism in the late 1960s.


  • to the dissemination of a method that can not only be useful in situations where traditionalregression is not appropriate, but also for comparative cases at the aggregate level where there isno suitable comparison group. The synthetic control method can be very useful to analyze theimpact of policies and events on specific outcome factors. An important caveat is that making useof this method requires caution, as this paper will show. Yet after addressing these challengesand estimating the many robustness checks necessary, one can arrive at an otherwise inaccessibleestimation with the use of a much better comparison unit. Brazil might have been the maindestination of FDI in Latin America for the past couple of decades despite not enacting a singleBIT, yet this does not mean it was achieving its full potential as FDI magnet. The analysis in thispaper suggests BITs would have boosted Brazils attractiveness and contributed to an even greaterinflux of FDI into the country.

    II. BITs and FDI

    The primary purpose of signing BITs is widely believed to be a promotion of the flow of FDI,as stated in preambles of thousands of existing BITs (Neumayer & Spess 2005). Dolzer (2005)goes even further and connects them to the necessary condition for the host state to achieveeconomic progress, and defines them in more details as the rules agreed upon by two parties"that serve to attract foreign investment by reducing the space for unprincipled and arbitraryactions of the host state and thus contribute to good governance" (Dolzer 2005, p. 954). Allee &Peinhardt (2011) argue BITs reassure fearful investors in two ways: first, by signing BITs, hostgovernments signal their underlying intention not to interfere with foreign investment; second, ahost governments credibility to abide by the terms of a contract increases when it enters into aBIT, which then stimulates greater FDI inflows (Allee & Peinhardt 2011). As Salacuse & Sullivan(2005) highlight, developing countries celebrate BITs with their developed counterparts thinkingof their subliminal grand bargain: "a promise of protection of capital in return for the prospect ofmore capital in the future"(Salacuse & Sullivan 2005, p. 77). As a result, countries have signedmore than 2,500 bilateral investment treaties, with provisions in them that allow foreign investorsto initiate arbitration proceedings against the host state. Most of the literature on BITs employsome type of credibility-based logic to argue that adopting these treaties should increase FDI flowinto signatories 3

    There is a growing consensus in the literature that signing BITs lead to more FDI flows4

    What enhances even further the validity of these findings is the fact that these studies examine anincreasingly wide range of countries over a long time-series and employing, arguably, more reliablecountry level FDI data. The biggest problem in the literature though is the lack of consensus onhow to empirically model FDI patterns due to the unclear nature of what determines FDI. Thereis little consistency in the literature in the covariates that are postulated by each scholar to explainworldwide FDI patterns. Therefore, teasing out the effect of BITs on FDI flows becomes a difficulttask dictated by theoretical priorities and preferences that consequently erode some of the validityof these studies as a whole. This is one of the main limitations of existing empirical approachestrying to understand the relationship between BITs and FDI. Although many studies employnuanced theoretical arguments, they often use a common, undifferentiated empirical strategygenerally focused on regression models involving specific sets of covariates determined by theresearcher. This paper suggests a different strategy, a different angle from which to tackle this

    3see Bubb & Rose-Ackerman (2007), Bthe & Milner (2009), Egger & Pfaffermayr (2004), Haftel (2010), Hallward-Driemeier (2003), Kerner (2009), Salacuse & Sullivan (2005).

    4 See Bthe & Milner (2009), Egger & Pfaffermayr (2004), Haftel (2010), Neumayer & Spess (2005), Salacuse & Sullivan(2005).


  • issue. Instead of building an empirical model controlling for novel covariates, the paper suggeststaking advantage of a unique situation in the international investment system and using a newand creative empirical tool; the Synthetic Control Method.

    Some researchers5 do not clearly see the causal arrow in the relationship going from BITs toFDI 6 Sauvant & Sachs (2009), for instance, suggests the connection between BITs and FDI is weakat best while Walde (2000) argues that there is no tangible evidence to "demonstrate investmentflows and a link to investment treaties." (Walde 2000, p.12). This strand of the literature tend tohighlight Brazils healthy FDI inflows in the 1990s and its antagonistic relationship with BITs asan example that the causal arrow might flow the other way, or not at all 7. Kalicki & Medeiros(2008), for instance, argue that Brazils position towards BITs reflects in part a confidence thatthese international instruments are not necessary as Brazil has led Latin America in FDI inflowsfor several years. Furthermore, Walde (2000) has highlighted Brazils situation as an example thatthe treaties might actually be negatively correlated to investment flows.

    The fast advancement and concurrent increase in government efforts to attract FDI by signingBITs during the 1990s makes the investigation of causality between the two both fascinating andchallenging. The international investment system, the history of BITs and Brazil presents us witha unique opportunity to use the synthetic control to investigate the impact of BITs on FDI in oneof the very few cases where the relationship between both does not go hand in hand. Most ofthe literature on BITs and FDI adopt a nonexperimental, ex post facto design in which they all useexisting data to examine existing relationships. This means these researchers would all have tobe careful with the time-order ambiguity, spuriousness and selection bias in their studies as theyall rely on after-the-fact econometric measures to counteract the lack of a concrete treatment.Most of the studies in the literature8 use a fixed-effects estimation to handle the time-invariantunobserved and uncontrolled-for factors that are correlated with the BITs and FDI inflows. Inthis sense, since the fixed-effects model includes a dummy variable for the units of analysis andeach point in time, the papers estimate a counterfactual in two ways. First, these analyses alluse a countrys FDI level prior to the implementation of a BIT as a counterfactual to estimate theimpact of BITs on such levels. Second, they also use the units of analysis as a comparison betweenthemselves in the fixed-effects model by including a dummy variable for each country to accountfor the possibility of persisting differences in FDI levels across countries. The fixed-effects modelis intended to remove the effect of the time-invariant characteristics so they can assess the neteffect of the predictors on FDI inflows. As comparison studies between countries that sets out toanalyze the numerous and sometimes elusive determinants of FDI flows, they all eventually facethe threat of ommited variables to the validity of their papers. As quasi-experimental designs thatuse the units as their own control over time, they all face the threat of history9. The change, orlack thereof, may be then accounted for by some "change-producing force of history" (Mohr 1995).As fixed-effects analyses that use the units of analyses as comparison groups between themselves,they all potentially face threats to internal validity. Although they use fixed-effects exactly tocontrol for the time-invariant initial differences between the units, all non-experimental designswith some form of comparison groups are subject to this threat. In a way, since we have thecountries as comparison groups between themselves, the threat of history identified above turnsinto a divergent history threat, meaning, "divergence in experience concerning these threats" (Mohr

    5see Sauvant & Sachs (2009), Yackee (2007), Walde (2000), Peterson et al. (2004), Motta Veiga (2004)6Aisbett (2007), for instance, suggests that the positive impact of BITs on FDI growth in some of the earlier studies is

    due to a failure to take into account certain methodological challenges.7see Walde (2000), Kalicki & Medeiros (2008), Motta Veiga (2004)8see Hallward-Driemeier (2003), Rose-Ackerman & Tobin (2005), Neumayer & Spess (2005), Salacuse & Sullivan (2005)9A history threat, as defined by Mohr (1995), is when something else accounts for all or part of an observed change

    over time.


  • 1995, p. 73). The choice of fixed-effects models can be a great econometric remedy to mitigatesome of the impact of selection bias and divergent history, yet in the end they are only unreliableband-aids since there are many other factors involved in the bilateral relationship between BITsand FDI that we cannot expect to know about. What this paper suggests is a method with whichwe can build a better comparison group and bring the analysis from a non-experimental designinto one closely resembling an experimental design. The synthetic control method is the tool thatwill allow us to carry out this task.

    As pointed out by Lemos & Campello (2015), the overwhelming change in strategies adoptedby governments toward foreign direct investment is better illustrated by Latin America in the1990s. They contend that the region had been strongly influenced by the Calvo Doctrine, whichthey say "determined that foreigners should receive the same treatment as domestic investors,and should therefore be subject to national regulation and courts" (Lemos & Campello 2015).Although, initially, this influenced the region not to sign the International Court on the Settlementof Investment Disputes (ICSID), eventually all countries in the region signed up to ICSID withthe exception of Brazil. Yet initially there was a suspicion that, although reticent, Brazil wouldnot stay completely out of the international investment agreement system since it signed 14 BITswith different countries, as shown on table 1. These BITs never saw the light of day though asthe Brazilian congress never ratified them10. Until now, we did not have the opportunity to reallytest Brazils position towards BITs in the FDI inflows into the country. It was only assumed bythe countrys politicians at the time that BITs would not make a difference in the already massiveamounts of FDI inflows into the country during that period.

    Table 1: Brazils signed BITs


    Country Date Signed

    Portugal 02/09/1994Chile 03/22/1994United Kingdom 07/19/1994Switzerland 11/11/1994France 03/21/1995Finland 03/28/1995Italy 04/03/1995Denmark 05/04/1995Venezuela 07/04/1995South Korea 09/01/1995Germany 09/21/1995Cuba 06/26/1997Netherlands 11/25/1998BLEU (Belgium-Luxembourg Economic Union) 01/06/1999Source: UNCTAD - Investment Policy Hub

    10See Haftel (2010) for a look into the importance of the ratification of BITs.


  • i. Brazil and BITs

    When it comes to policy-making in Brazil, and especially in international affairs, presidents arethe most influential factor in the system. In the particular case of international treaties, theconstitutional prerogative of negotiating them falls to the president, but for treaties to enterinto force they still need congressional approval 11. Within the executive branch, the ForeignAffairs Ministry, known as the Itamaraty, is in charge of the delegations tasked with negotiatinginternational treaties. The historical changes Brazil went through in the 1990s as well as thechanges in presidents and parties in power during this time are important to understand thefate of BITs in the country. The first popularly elected Brazilian president after the conclusionof the military regime was Fernando Collor de Mello. When he came to power in 1990, heestablished largely neoliberal economic policies. His short presidency ended in an impeachment,yet his successors, Itamar Franco and Fernando Henrique Cardoso, continued such policies. Thesepolicies, driven by the neoliberal turn, were aimed at attracting and protecting investment andwere promoted largely because of the massive wave of privatization done by these administrationsin the 1990s. The opposition was vehemently against these neoliberal policies not only politicallybut, most importantly, ideologically. The BITs Most Favored Nations (MFN) clauses also madethe opposition unconfortable because it put Brazils sovereignty in jeopardy by offering preferredterms to foreign investors while taking away Brazils ability to decide how and with whom tonegotiate agreements. Also, the BITs compensation provisions for expropriations were, accordingto the opposition, in conflict with the preexisting constitutional structure to deal with such matters(Welsh et al. 2014, p. 119). The opposition it was facing started to come not only from thevociferous Workers Party anymore, but also from the ruling coalition. Even worse, Lemos &Campello (2015) contend that, as a policy bearing potentially diffuse benefits, "there was never aclear constituency pushing for the treaties ratification" (Lemos & Campello 2015, p. 25)12. Thisseemed to further drain the Presidents willingness to fend for the BITs.

    By 2002, when the BITs had absolutely no more chance of being ratified13 the then Ministerof Foreign Affairs, Celso Lafer, recognized the goal of BITs was to strengthen the normativeframework for foreign investment with the aim of reinforcing Brazils position as a resourcedestination, in line with the international environment at the time (BRASIL 2002, p. 54409 54415).Yet the Minister also highlighted the perception that the overarching protective devices for theforeign investors would operate in detriment to the nations jurisdiction and society (BRASIL 2002,p. 54409 54415). He put forward then the position most opposing politicians in Brazil held atthe time, that the inexistence of BITs has not affected the countrys position as an important FDIdestination. Brazilian officials argued this was due to Brazils stable legal rules in the domesticlevel and the strength of Brazils economy since 1994 (BRASIL 2002, p. 54409 54415). Theposition of the Foreign Affairs Minister was a reflection of the ebbing support for BITs by theadministration and was understandably the last nail in the BITs coffin. Even though the Brazilianexecutive was historically considered very powerful, the office proved, after a negotiation thatlasted almost a decade, completely unable of enacting a single BIT (Lemos & Campello 2015).This scenario led President Fernando Henrique Cardoso to withdraw in 2002 all BITs from theBrazilian Congress two weeks shy of handing over the presidency to Luis Incio Lula da Silva.Lula, the opposition candidate, acceded to power on January 1st, 2003. Having the Workers Partyin power did not make the future of the BITs and international investment protection in Brazil

    11After the treaty has been negotiated, the President sends them to Congress for consideration. Congress has twochambers and both have to be involved in the ratification process.

    12This is consistent with other research on the lack of corporate lobbying for BITs13Treaties have to receive simple majorities in floor votes in the Chamber of Deputies and then the Senate. See Welsh

    et al. (2014) for a detailed look into this process.


  • look any better.This situation left Brazil as the potential shining example that a major host country does not

    need BITs if it has a strong economy and proper domestic protection to foreign investors14. Again,is this really the case? This paper will try to answer this question and the ones posed above not bylooking at Brazils constitutional provisions 15 or its economic characteristics at the time. It willdo so by estimating the impact of its unique position towards BITs on the FDI inflows into thecountry at the time with the help of the synthetic control method.

    III. Methods

    In a comparative study such as the one proposed here, the unit of analysis is an aggregate entity,and, as such, there is not a single comparison unit that is suitable for the task. There are no twocountries that resemble each other in every factor that impacts FDI inflows into them. Even if wefind countries that resemble the unit of interest (Brazil) in some aspects, there will be a myriadof other factors that will differ considerably enough to render the comparison dubious. Suchheterogeneity between countries may create problems when comparing the experiences of differentcountries and interpreting the results. What the Synthetic Control offers is an opportunitty togather a number of comparable units and use a data-driven method to generate a synthetic versionof the unit built from a weigthed average of of all potential comparison units that best resemblesthe characteristics of the case of interest (Abadie et al. 2015, Abadie & Gardeazabal 2003, Abadieet al. 2010). In this case, the comparison units are intended to reproduce the counterfactual of thecase of interest in the presence of the event or intervention under scrutiny (Abadie et al. 2015).An important aspect of this paper is that it is defined instead by the occurrence of a sequence ofevents or interventions and otherwise similar but affected units. Instead of estimating the impactof an intervention on an affected unit, we will flip the roles and estimate here the impact of anunaffected unit against otherwise similar but affected units.

    There is also not one single BIT that affected all units at one point in time, but the BITs historyalso awarded us with another opportunity as they have an identifiable period in time where theyexploded in the global scenario. Usually, most studies that use the synthetic control investigatethe impact of a policy or intervention that happens at a particular point in time16. Having a singleintervention happening at a specific point in time is not vital though if we are studying an eventthat has an identifiable starting point. For instance, in the very first study done with syntheticcontrol, Abadie & Gardeazabal (2003) investigate the economic effects of conflict using the terroristconflict in the Basque Country as a case study17. In this case, they do not have a specific yearwhere the conflict started as it was a progressive sequence of events18. As Abadie & Gardeazabal(2003) show, the outset of terrorism in the Basque Country cannot be traced back to a specific year,yet we can trace it to a specific period in time and pin it down in a given year. This is what willbe done here as there is no specific starting date or single BIT being analyzed, but, as figure 1shows, there is a clear point in time where BITs take off in the world, 1990. This study will use acombination of other countries to construct a synthetic Brazil which resembles relevant economiccharacteristics of this country before the outset of BITs in the early 1990s.

    14see Lemos & Campello (2015) and Kalicki & Medeiros (2008)15For a detailed look on this subject, see Lemos & Campello (2015) and Kalicki & Medeiros (2008)16See Abadie et al. (2010).17They use a combination of other Spanish regions to construct a synthetic control region which resembles relevant

    economic characteristics of the Basque Country before the outset of Basque political terrorism in the late 1960s.18For instance, ETA was founded in 1959, however, it was not until 1968 that ETA claimed its first victim, but it did not

    implement large-scale terrorist activity until the mid-1970s (Abadie & Gardeazabal 2003).


  • Figure 1: The global surge in BITs signed between countries

    The choice of countries that will comprise the donor pool is a vital part of the syntheticcontrol method since the comparison units are meant to approximate the counterfactual. Thereare over two hundred countries in the world, yet not all have enforced BITs or have levels of FDIinflows large enough to be considered in this study. This paper will implement systematic roundsof requirements necessary for countries to be considered and use only the ones left at the end forthe donor pool, as shown on table 2 below.

    Table 2: Donor Pool Selection Criteria

    Criterion-based Selection Rounds

    Round Number Criterion

    1 FDI Host Economies2 Meaningful FDI Inflows3 No BIT by 19954 BIT by 1980

    First and foremost, the comparison units have to be susceptible to the treatment the same wayas the treated unit. This means the first round has to eliminate FDI home countries, meaningcountries that are major exporters of capital and usually initiators of a treaty. Since we are settingthe treatment period to 1990, every country that was an OECD member before that is automatically


  • excluded from the pool. This round obviously eliminates most of the capital rich economies in theworld. Countries also need to have a significant level of FDI inflow prior to 1990 because this isthe vital pretreatment period in which the weights in the predictor variables and countries willbe calculated. To this end, every country with less than U$ 10 million yearly in FDI inflows onaverage prior to 1990 will also be excluded from the pool. This round eliminates small countries,island states and countries that were not open to foreign investment prior to 199019. Next, weneed to consider countries that have been affected by the treatment, meaning countries that have aBIT in force by at least 1995. The logic behind this round is that, after 1995, the impact of the BITis not going to matter for our crucial treatment period. This eliminates important countries suchas Colombia, which only had a BIT enforced by 2004, Iran, which enforced its first BIT by 1997,and Saudi Arabia which did it by 1996. A crucial period for the estimation of the synthetic Brazilis the decade right before the treatment period, 1990. Although the explosion of BITs happenedin the early 1990s, there are a few countries that have enforced a number of BITs prior to that.We will then exclude countries that had any BITs in force by the beginning of the 1980s so as notto taint the pretreatment with any impact BITs might have had in these countries. Some of thecountries excluded by this round include Paraguay, which had 2 BITs in force by the beginning ofthe 1980s; South Korea and Indonesia both had 7 BITs in force by that time; and Egypt, which hada staggering 9 BITs in force by the end of the 1970s. After filtering all the countries by the stepspreviously mentioned, we were left with 27 countries in our donor pool, as the table below shows.

    Table 3: List of Countries in the Donor Pool.

    Argentina Peru CameroonBolivia Uruguay CongoChile Venezuela GhanaCosta Rica Jamaica NigeriaEcuador Cyprus ChinaEl Salvador Israel South AfricaHonduras Sri Lanka TurkeyMexico Hong KongNicaragua IndiaPanama Philippines

    The task of choosing the predictor variables that will contribute to the construction of thesynthetic Brazil is no simpler than developing the donor pool. We have to select factors thatinfluence the decisions of investors when choosing a country to park their funds. This is no smalltask as empirical studies of determinants of FDI show substantial differences in specificationswith little agreement on the set of included covariates. This paper will use Blonigen & Piger(2014) study on FDI determinants to guide its choice for predictor variables. They use Bayesianstatistical techniques that allow one to select from a large set of candidates those variables mostlikely to be determinants of FDI activity (Blonigen & Piger 2014)20. Brazil has to be given specialattention in the selection of predictor variables as we are trying to reproduce specifically the FDI

    19This means all the countries that were a member of the Soviet Union are not going to be considered for the donor pool.20This will not be the only position considered though because, first and foremost, their study is for bilateral FDI and,

    as such, some of the determinants are specifically related to that. For instance, most of the traditional gravity variablessuch as proximity, remoteness, common language or colonial relationships, are not applicable to a study on aggregate FDIinflows


  • inflows into this country. We have to consider what made Brazil so attractive to investors in the1990s and weigh the countries in the donor pool accordingly. In the midst of uncertainties onFDI determinants in the literature, there are some factors in which there are consensus and areconfirmed by Blonigen & Piger (2014) study. The main gravity variables were the only ones withone hundred per cent chance of inclusion in Blonigen & Piger (2014) study and as such will all beinlcuded here. One of them is population size, which is usually a proxy for country size or marketsize and it should definitely be part of a study considering Brazil. GDP per capita should alsobe considered, which is usually a proxy for the level of development of a country. GDP growthis also included as it is often considered a proxy for economic development, a commonly heldrobust predictor of FDI. We also consider the urban concentration of a given country as this hasalso a significant inclusion probability in Blonigen & Piger (2014), although to a lesser extent.Other variables with high inclusion probabilities considered in this paper are trade openness, legalinstitutions and skill level.

    As previously seen on this paper, one of the reasons given by Brazil for it to be a majordestination for FDI despite not having enacted a single BIT is the strong protections it givesto foreign capital21. A government must secure private property rights not only for a merketeconomy to work, but, most importantly, to assure foreign firms their property rights will berespected22. As Wellhausen (2014) argues, foreign firms lack the protection to their property rightsthat domestic firms have, since foreign firms are not part of the social contractbetween citizensand the state. In light of this, the level of property rights protection will also be considered bythis paper. For this variable we use the ICRGs23 dataset to create a measure of property rightsprotection which aggregates four indicators of the guide; investment profile, bureaucratic quality,corruption, and law and order from the PRS PRS-Group (2007) data set on country risk24. Theonly difference is we break down the measure into its original single parts to better capture thelevel of legal institutions suggested by Blonigen & Piger (2014). This will allow us also to useinvestment profile as a proxy for skill level and keep the specification as parsimonious as possible.The main GDP variables as well as population and trade statistics come from the World BankDevelopment Indicators. UNCTAD is the source for FDI data.

    IV. Results

    The sample for this paper is a strongly balanced longitudinal data set where all units are observedat the same time periods, t = 1, 2, ...T. The dataset also includes sizable pre- and post-interventionperiods, from 1970 to 2010, giving us valuable 20 years for each of both periods. Abadie et al.(2015) define the synthetic control as a weighted average of the units in the donor pool that can berepresented by a (J 1) vector of weights W = (w2, w3, ..., wJ+1), with 0 6 wj 6 1 for j = 2, 3, ..., Jand w2, w3, ..., wj+1 = 1. They propose "selecting the value of W such that the characteristics ofthe treated unit are best resembled by the characteristics of the synthetic control" (Abadie et al.2015). In this case, let X1 be a (k 1) vector containing the pre-BITs characteristics of Brazil thatwe strive to match as closely as possible, and let X0 be the k J matrix collecting the values ofthe same variables for the other countries in the donor pool. The weights are chosen so that thesynthetic Brazil most closely resembles the actual one before the BITs wave swept the world in theearly 1990s. The difference between the pre-BITs characteristics of Brazil and its synthetic versionis given by the vector X1 X0W. The method selects the synthetic control, W, that minimizes

    21For a detailed look into this subject, see Lemos & Campello (2015) and Kalicki & Medeiros (2008)22see Wellhausen (2014)23International Country Risk Guide, PRS Group 2007.24Measure also used by Allee & Peinhardt (2011) and Jakobsen & De Soysa (2006)


  • the size of this difference, as follows:



    m(X1m X0mW)2, (1)

    where X1m is the value of the m-th variable for the treated unit, X0m is the 1 J vectorcontaining the values of the m-th for the other donor units in the pool and m is a weight thatreflects the relative importance assigned to the m-th variable when the discrepancy between X1and X0mW is measured (Abadie et al. 2015, Abadie & Gardeazabal 2003, Abadie et al. 2010). Thematching variables in X1 and X0, which are not affected by the intervention (BITs), are meantto be predictors of post-BITs outcomes. The vector W is chosen to minimize some distance, X1 X0W , between X1 and X0W, subject to w2 >, ..., wJ+1 > 0, w2 + ... + wJ+1 = 1. Thesynthetic control method measures the discrepancy between X1 and X0W, by employing:

    X1 X0W v=(X1 X0W)V(X1 X0W), (2)

    where V is some (k k) symetric and positive semidefinite matrix (Abadie et al. 2010). Thevariables with a large predictive power on FDI inflows for Brazil, should be assigned large mweights , which is done through the cross-validation method below:

    Y1t j+1


    wj Yjt, (3)

    where Yjt is the outcome of unit j at time t, Y1 is a (T1 1) vector collecting the post-BITsFDI values for Brazil, Y1 = (Y1T0+1, ..., Y1T), Y0 is a (T1 J) matrix where column j contains thepost-BITs values of FDI inflows for unit j + 1. The synthetic control estimator of the effect of nothaving BITs in force in Brazils FDI inflows is given by the comparison of post-BITs outcomesbetween the actual Brazil, which has no BITs in force, and the synthetic Brazil, which has BITs inforce. In other words, the goal is to approximate the FDI inflows pattern that Brazil would haveexperienced had it enforced treaties with other countries. This counterfactual pattern is calculatedas the FDI inflows into the synthetic Brazil, Y1 Y0W. Meaning, for a post-BITs period t (witht > T0) the synthetic control estimator of the effect of no BITs in force, the treatment, is given,aacording to Abadie et al. (2015), by the comparison between the outcome for the treated unit,Brazil, and the outcome for the synthetic control at that period, as seen above. This minimizationof the difference between the pre-BITs characteristics of Brazil and its synthetic version is subjectto the constraints that the weight assigned to each unexposed country should lie between zeroand one and that the sum of the weights is bounded by one. The weights are assigned to countriesin the donor pool in such a way that the pre-BITs FDI inflows and the covariates that are thoughtto influence FDI inflows are comparable to those of Brazil before the surge in BITs in 1990. Thiscomparability is determined by the minimization of root mean square prediction error (RMSPE)in the pre-BITs period, which measures the lack of fit between the trajectory of the outcomevariable and its synthetic counterpart. After carefully selecting the units in the donor pool and thepredictor variables used to weigh them, we run the model to obtain the synthetic estimation ofthe outcome variable. Below is the table with the weights of the countries that contributed to thesynthetic Brazil as well as a visual representation of the FDI inflows in the actual and syntheticBrazil.

    As we can see on figure 2, the synthetic Brazil has a consistent growth in FDI inflow immediatelyafter the treatment, whereas that surge happens around five years later for the actual Brazil. Wecan also see that around 1997, the FDI inflow to Brazil surpasses the inflow to its synthetic version


  • Table 4: Composition of Synthetic Brazil

    Countries that Contributed

    Country Weight (Percent)

    Argentina 19China 29Hong Kong 1Israel 1Mexico 35Venezuela 15

    Figure 2: Results of the Synthetic Control Estimation

    and stays above it for a period of three to four years. This is the only period Brazils actual inflowovertakes the synthetic estimation for the whole period between 1990 and 2010. This requiresfurther investigation and we will talk about this later on this paper. Initially, what we can see isthat the synthetic version remains for the most part above the actual FDI inflow for Brazil in thisperiod suggesting BITs could have augmented foreign investments in the country.

    We can now estimate the impact of Brazils exclusion from the BIT system on the FDI inflowsinto the country using the synthetic Brazil. Figure 3 better illustrates this effect and how the


  • estimation will be calculated. Everything below the dotted line, which represents the syntheticestimation, is a negative effect on FDI inflows. The impact is calculated as such:

    Y1t J+1


    wj Yjt (4)

    Where Yjt is the FDI inflow of unit j at time t; Y1 is a (T1 1) vector collecting the post-BITsvalues of FDI inflows for Brazil. The synthetic control estimator of the effect of not having a BITin force is given by the comparison of postintervention FDI inflows between Brazil, which doesnot have a BIT in force, and the synthetic control, which is exposed to BITs, Y1 Y0W. So, fora postintervention period t (with t > T0), Abadie et al. (2015) argue that the synthetic controlestimator of the effect of the treatment is given by the comparison between the outcome for thetreated unit and the outcome for the synthetic control at that period as shown above. The analysisallows us to estimate that Brazils reluctance to enact BITs had a negative impact of nearly 90billion US dollars 25 in the two decades between 1990 and 2010.

    Figure 3: Effect of Brazils Exclusion from the BIT System

    Another aspect in the visual analysis that stands out and needs further investigation is thesudden spike in FDI inflows in Brazil in the late 1990s. Figure 4 shows this is the only portionof the actual Brazil that goes above the synthetic version. During this period, Brazil had anunusually high inflow of FDI due to the large scale privatizations carried out in this period by

    25US$ 89, 694, 803.70 to be more precise


  • the government. According to the World Bank, as the bulk of the energy and telecommunicationcompanies were being privatized between 1997 and 2000, FDI into Brazil peaked at U$ 33 Billion(World Bank 2004). Privatizations proceeds then plummeted in 2001 to U$ 2 Billion from anannual average of U$ 19 Billion between 1997 and 2000. The World Bank argues a significantfactor in this sharp decline in 2000 in FDI inflows is due to the winding down of the large scaleprivatizations. We can see on figure 4 that the World Banks account matches exactly what the FDIinflow data says. Before analyzing these results any further and taking any conclusions, we haveto run placebo experiments and robustness checks to check how confident we are in the methodand model employed.

    Figure 4: Privatization and FDI Inflows to Brazil

    V. Placebo and Robustness Checks

    As placebo experiments, Abadie et al. (2015) suggest two types: the in-time and in-space steps.The in-time placebo is only possible when there is enough pre-tretament data and it suggeststhe treament be moved back to a point in time where the treatment was not implemented yetand analyze the consequences immediately after. The in-time placebo shown in figure 5 showsthe impact of the intervention is specific to the time selected. Our confidence about the validityof this result would dissipate, as argued by Abadie et al. (2015), if the synthetic control methodalso estimated large effects when applied to dates when the intervention did not occur, which inthis case we set to 1980. As we carefully added to the donor pool selection criteria, no country


  • in the donor pool had a BIT in place by 1980, so if we had substantial effects in the level of FDIinflow in the synthetic unit in the 1980s our confidence that the intervention was the cause ofthe significant effects in the 1990s would be severely undermined. This in-time placebo was onlyfeasible because there was enough available data for a sufficiently large number of time periodswhen no structural shocks to the outcome variable occurred (Abadie et al. 2015). In this case therewas enough data for the outcome variable as well as the main predictor variables26 going backto 1970, which allowed the placebo treatment period to be set to 1980. As the figure (5) belowshows we have no significant movement in the outcome variable for the period after the placebotreatment period and before the actual treatment period.

    Figure 5: In-Time Placebo Study

    In the interpretation of the in-space placebo we also have to reverse the logic of the explanationgiven by Abadie et al. (2015). Acoording to Abadie et al. (2015), the alternative model of inferenceobtained with the synthetic control inspires confidence because, if we obtained estimated effects ofsimilar or even greater magnitudes in cases where the intervention did not take place, the impactof the intervention under scrutiny would be severely undermined. In the case of this study, wereverse the roles of treated and untreated units but can still tease out the effect of the treatment(BITs), we just have to remember that some of the logic of large effect on the treated unit issuppose to be reversed also. Instead of obtaining synthetic control estimates for countries that didnot experience the event of interest, we do it for the countries that did experience it. Applying

    26The only variable excluded is property rights due to the fact that the ICRG dataset does not go as far back as 1970.


  • this idea to each country that contributed to the synthetic Brazil in the donor pool27 allows us tocompare the estimated effect of Brazil not having any BIT in force to the distribution of placeboeffects obtained for other countries that have BITs in place. In this case, if there are other effectswhich are of lesser magnitude, we would seriously doubt the negative effect of the absence of BITson Brazils FDI inflows. The purpose is to asses whether the gap observed for Brazil may havebeen created by factors other than BITs. As we can see on figure 6, the in-space placebo showsus that the synthetic Brazil seems to have the smallest effect, or at least there are no other effectsmore extreme then Brazils. This is so even with the unusual shock, which we will discuss later,between 1997 and 2000 which raises Brazils FDI inflows significantly.

    Figure 6: In-Space Placebo Study

    The in-space placebo experiment (figure 6) serves as an illustration that the gap betweensynthetic and actual Brazil is not merely an artifact of the inability of our analysis to reproducethe FDI inflows in the absence of BITs. To further address this question, we follow (Abadie &Gardeazabal 2003) and perform a single placebo experiment applying the method that we used tocompute the gap for Brazil to a single country that actually has BITs in place. Following (Abadie &Gardeazabal 2003), we select first a country with a large weight in the synthetic control for Brazil;Argentina. In addition to being a big contributor to the synthetic Brazil, Argentina is also Brazilsneighbor and, as such, influenced the most by geographical proximity as well as it underwentprivatizations in the same period. Argentina also had a significant amount of BITs in place duringthe 1990s, so it also is a good example to contrast with Brazil. Figure 7 shows the actual level of

    27Argentina, China, Hong Kong, Israel, Mexico and Venezuela


  • FDI inflows and the synthetic estimations of both Brazil and Argentina. We can see on figure 7 thatthe synthetic version closely reproduces the FDI level of Argentina while significantly changingsynthetic Brazil. If the synthetic version of Argentina were to be significantly different then theactual Argentina, both with BITs in force, this would put the results of the method and its capacityto reproduce the synthetic Brazil in question.

    (a) Actual Argentina (b) Synthetic Argentina

    Figure 7: Actual and Synthetic Argentina and Brazil

    We will select now, as a second single placebo study, a case that will help in better visualizinghow the method is able to tease out the effect of Brazil never enacting BITs. There are only twocountries in the donor pool that have higher FDI inflows than Brazil and they both contributeto the composition of the synthetic unit; China and Hong Kong. The idea is to use one of themnow as the placebo case so we can better visualize the ability of the synthetic control method inreproducing the effect of the treatment on the outcome variable. There are also only two countriesin the donor pool that have an external shock in the FDI inflows around the same time as Brazilslarge scale privatizations. One of these countries has already been used before, Argentina, and theother is part of the two countries with higher FDI inflows previously mentioned: Hong Kong. Asyou can see on figure 8, Hong Kong had a massive spike on the FDI inflow in 2000, much biggerthan Brazils albeit much shorter in duration. This is in part due to Chinas much awaited entryinto the WTO in 2001. According to UNCTAD (2001), MNCs planning to invest in mainland Chinahave "parked" funds in Hong Kong, in anticipation of Chinas entry into the WTO. Hong Kongalso had a major cross-border merger and acquisition (M&A) in telecommunications in 2000 thatcontributed heavily for the spike we see in the figure below (UNCTAD 2001). This spike in FDIinflows around the same time as Brazils and its higher volumes of FDI inflows makes Hong Kongthe perfect candidate for the second placebo study that will help us examine the synthetic controlmethods capability of reproducing the treatment on the outcome of interest in Brazil. As youcan see on figure 8, the synthetic control reproduces very well the actual FDI inflow into HongKong, with BITs in force, even with the external shock previously mentioned, while significantlychanging Brazils FDI inflow pattern.

    We can also, as suggested by Abadie et al. (2015), run robustness checks to test the sensitivityof our main results to changes in the country weights. The first robustness check will have to dowith the choice of countries to the donor pool. The selection process for the original donor poolfor this paper was carried out as systematically as possible so as to avoid issues with selection bias.Yet there were some countries in the donor pool that might have be considered as comparable


  • (a) Actual Hong Kong (b) Synthetic Hong Kong

    Figure 8: Actual and Synthetic Hong Kong and Brazil

    to Brazil to start with, such as Jamaica or Cyprus. This first robustness check is to run the samemodel but with the pool consisting only of populous developing countries. This choice wouldhave been open to selection threats or open to the subjectivity the synthetic control method itselftries to overcome, but will be carried out here as a robustness check. For this step we will restrictour donor pool to Mexico, China, India, Argentina, South Africa and Turkey. As you can see onfigure 9, the synthetic estimation is not very different from the original one, yielding actually aneven bigger negative impact for Brazils position towards BITs. This synthetic Brazil is a weightedaverage of less countries, as expected, yet composed of countries that were also part of the originalestimation, such as Argentina (45%), Mexico (15%) and China (40%).

    In the next robustness check we go back to the original donor pool but we reduce the predictorvariables used to weigh them. This time we run the same model omitting trade openness and theICRG property rights variables; investment profile, bureaucracy, law and order and corruption.The idea is we keep only the gravity variables with very high inclusion percentages in Blonigen &Piger (2014) study and check how this very parsimonious model fares. Again, we can see on figure10 that the estimation does not change much and it also yields a bigger negative impact as theprevious one. Coincidentally enough, the synthetic Brazil is built from the same three countries asbefore but with different weights; Mexico (71%), Argentina (14%) and China (15%).

    The next robustness check builds on the logic of the previous one yet it expands the predictorvariables this time. The logic here is to check the divergence of the estimation if we actually ranthe model controlling for various extra factors. The literature is filled with different covariatesused by different researchers in their quest to control for FDI determinants. In this step wewill introduce five new covariates to the original model to control for exchange rate volatility28

    , financial openness29 , domestic political shocks30, domestic economic shocks31 and externalthreats32. As we can see on figure 11, this is the robustness check that changes the least from theoriginal model. Even if we add extra covariates, which were randomly chosen, we still have the

    28The method for creating this variable follows Li & Resnick (2003)29The method for creating this variable follows Chinn & Ito (2008).30This an unweighted count of the number of general strikes, government crises, antigovernment demonstrations, riots,

    purges, revolutions, or acts of guerilla warfare in a given country in a given year, from Banks & Wilson (2005) Cross-PolityTime Series data set.

    31Variable created with Laeven & Valencia (2008) dataset on systematic banking crises.32Variable taken from the PRS-Group (2007) political risk guide (IGRG).


  • Figure 9: Reduced Donor Pool Model

    same negative estimation on the FDI inflows into Brazil.For the last robustness check, we change the treatment period rather than the donor pool or

    predictor variables. As we saw before on this paper, Brazil signed the first of its ill-fated BITs in1994, so we will assign the treatment period to this date rather than the original 1990. Some mightargue 1994 would be a better treatment period since this is when Brazil first signed a BIT andwhen it should have been ratified and gone into effect. As we can see from figure 12, this does notseem to be the case as the synthetic Brazil starts having a significant difference from the actualBrazil prior to 1994. By 1994 we already have a significant number of BITs in the world signedbetween countries, hence the original logic of choosing 1990. Furthermore, we can see on figure 12that, regardless of the synthetic Brazil having already a significant difference before 1994, it doesstill resemble the same trajectory it had with the original model. The synthetic estimation in thisstep also shows the absence of BITs having a negative impact on FDI inflows into Brazil, albeitthis was the smallest one of all the robustness checks previously seen. This step also has morecountries contributing to the synthetic unit than the previous ones, with Argentina (37%), Mexico(24%), Venezuela (1%), Hong Kong (0.1%) and China (29%).

    The various placebo experiments and the robustness checks serve to increase the confidencein the model used and the synthetic estimation obtained. these various steps might corrode theconfidence in the particular amount estimated for the impact of the treatment, but it certainlyraises confidence in the negative impact on Brazils FDI inflows of the country not having enactedany BITs. As pointed out by Abadie et al. (2015), critics of Mills Method of Differences could


  • Figure 10: Reduced Specification Model

    rightfully point out that the presence of unmeasured factors affecting the outcome variable as wellas by heterogeneity in the effects of observed and unobserved factors may limit the applicability ofthe synthetic control method. This is the case specially in studies such as this one as there are manyuncertainties in the literature as to what determines the outcome variable, FDI inflows. However,using a linear factor model, Abadie et al. (2010) argue that if the number of preintervention periodsin the data is large, matching on preintervention outcomes helps control for unobserved factorsand for the heterogeneity of the effect of the observed and unobserved factors on the outcome ofinterest (Abadie et al. 2015). In other words, the difference in FDI inflows following the presence ofBITs is interpreted as produced by the BITs themselves, or lack thereof, once it has been establishedthat the actual Brazil and its synthetic version have similar behavior over extended periods of time.In this sense this study was able to remedy for the issues with the outcome variable by employinga particularly large number of preintervention periods in its analysis.

    VI. Discussion

    This paper helps to shed light into a puzzle that has intrigued the BIT literature and shows througha different pathway that BITs might have a more important role in attracting FDI than Brazilsmisleading case had otherwise shown. Brazil was attracting massive amounts of FDI in the 1990sand this led Brazilian officials to state the country did not need them. Contrary to the BIT wave


  • Figure 11: Increased Specification Model

    sweeping through the world in the 1990s, Brazil remained steadfast in its antagonistic positionand also managed to be the main destination of FDI in Latin America for a long period of time.Brazil has been seen as an exception to the increasingly widespread view that BITs lead to greaterFDI flows. Brazils notorious absence from the BIT system could also be seen as the product oftechnical and political barriers that have impeded the ratification of BITs as well as the historicaland ideological stance of the region. All the other countries in the region eventually joined the BITbandwagon, yet why was Brazil the only country that resisted the BIT fever? There is not onespecific reason, but, arguably, a combination of them and the right time and place for them totake hold. Brazils huge economic power, the positive economic outlook at the time, the lack ofpowerful enough opposing vested interest in combination with ideological and nationalistic viewswere the perfect ground for resistance to BITS to take hold. Yet, most importantly, in the absenceof a clear negative impact on FDI inflows, Brazil did not seem to have the crucial pressure othercountries had when considering entering into BITs.

    As the literature on FDI has shown throughout the years, investigating the impact of BITson FDI is anything but a simple task. Actually, investigating the impact of any single factoron the level of FDI into a country is bound to be a difficult task due to the myriad of forcesinfluencing it that feed off each other. This situation poses a major challenge for empirical analysesgrappling with omitted variable bias and multicollinearity due to the interconnection of all possibleexplanatory variables. A possible and interesting study to understand the impact of BITs on FDIin inflows into a given country would have to involve a better counterfactual. The lack of a proper


  • Figure 12: Different Treatment Period Model

    counterfactual and a study on the case where BITs and FDI do not go together is the hole thispaper is trying to fill and the main contribution of this paper to the international investmentliterature. The analysis on this paper suggests Brazil could have attracted more FDI between 1990and 2010 even when it was already a major destination for FDI despite not enacting a single BIT.We cannot use these findings as proof that other countries would have attracted large(r) amountsof FDI because of BITs. What the paper can be is an indication that one of the few countries in theworld that never had a BIT enacted could have attracted more had it enforced any. The method isnot meant for generalizations and is better employed in investigations of the impact of a particularevent or intervention in a particular country. In this case for example, it can be used to show thatBrazilian officials were right in assuming Brazil did not need BITs to attract FDI, but were wrongin assuming BITs would have no effect on the level of FDI into the country. They were right inthat the lack of BITs did not stop foreign investors pouring money into Brazil in that period, but asynthetic Brazil would have done better in these two decades had it had BITs in place.

    The synthetic control method has some shortcomings, or, better yet, challenges, that needto be addressed by the researcher to make the results as robust as possible. If one of the mainadvantages of the synthetic control method is to provide a systematic method as free of subjectivityin the choice of comparison group as possible, researchers need to pay special attention whenchoosing the donor pool and the predictor variables used in the model. There is no guidanceprovided in choosing the predictor variables when using the synthetic control method. If thechoice is not set before running the model it can lead to cherrypicking of the variables that yield


  • the best results. This is why this paper took the parsimonious route when choosing the predictorvariables used to claculate the country weights. This is why the choice of variables was alsoguided by Blonigen & Piger (2014) influential work on FDI determinants. A possible step that canbe taken to improve the choice of predictor variables when using the synthetic control method isto conduct, or use already existing, studies on the determinants of the outcome variable whenpossible33. Also, when running the robustness checks we can analyze what is the impact in theestimation of different specifications to the model as well as different compositions of the donorpool, as this paper showed. For the choice of units to the donor pool, we can always providecriteria which the units need to meet and use all the ones left at the end as donors, as done in thispaper. This can lead to some very disparate units being left in the pool that might confound thecalculations of the synthetic model. Some researchers might prefer smaller donor pools with unitsthat resemble the unit of interest as closely as possible. A useful next step in this sense then wouldbe to follow the criteria route and then estimate a propensity score to further restrict the donorpool to countries with similar observable characteristics prior to invoking the synthetic control.

    Brazil was riding a wave of privatizations and had a strong economy as a basis, yet had it hadBITs in place this could have been a larger and longer lasting wave. The papers estimation and itsrobustness checks show that, regardless of the composition of the donor pool and the choice ofpredictor variables, Brazil missed out on a significant amount of FDI inflows due to its positiontowards BITs. The question that remains is: was it a fair price to pay to keep its constitutionalsovereignty intact? The fact that Brazil, as an increasingly stronger FDI exporter, is now pushingfor its own brand of BITs 34 might say otherwise. Even more interesting is that Brazil, that wasswimming against the tide before by being one of the very few countries never to enact a BIT, isagain swimming against the tide now as its push comes at a time when a growing number ofcountries are abrogating BITs 35.

    We were granted a unique opportunity due to the nature of the explosion in BITs in the 1990sand Brazils exclusive position towards them to use a synthetic control method to estimate theimpact of such position. The synthetic control method is a great tool, yet we need caution whenusing it so as not to fall in the same problems we try to avoid when using it in the first place. Thispaper helped us pull the veil on the puzzle that was FDI into Brazil in the 1990s that had intriguedthe literature for some time. It enabled us to evaluate the relationship between BITs and FDI froma different angle and in the process reinforce the view that BITs lead to greater FDI from one ofthe few cases in the world where they do not go hand in hand. It also allowed us to estimate theimpact of Brazils position toward BITs and better analyze the arguments it used to defend it. Thecombination of the right place at the right time allowed Brazil to remain steadfast against BITswhile still attracting massive amounts of FDI. Brazil was at a table, voraciously eating from theFDI banquet even without bringing the BIT invitation as the other guests. Little did they knowthat there was a bigger table, more plentiful and that it would have been there had it brought theBIT invitation to the party.

    33In the case of this study, since we want to reproduce the effect on the outcome variable on only one specific country,we could perform an impulse response function on the possible determinants to analyze which ones meaningfully impactFDI inflows.

    34The Cooperation and Investment Facilitation Agreement (CIFA). For a detailed look into CIFA, see Gabriel (2016),Brauch (2015), Morosini & Ratton (2015)

    35see Peinhardt & Wellhausen (2016)


  • VII. References


    Abadie, A., Diamond, A. & Hainmueller, J. (2010), Synthetic control methods for comparative casestudies: Estimating the effect of californias tobacco control program, Journal of the AmericanStatistical Association 105(490), 493505.

    Abadie, A., Diamond, A. & Hainmueller, J. (2015), Comparative politics and the synthetic controlmethod, American Journal of Political Science 59(2), 495510.

    Abadie, A. & Gardeazabal, J. (2003), The economic costs of conflict: A case study of the basquecountry, The American Economic Review, 93(1), 113132.

    Aisbett, E. (2007), Bilateral investment treaties and foreign direct investment: correlation versuscausation, Department of Agricultural & Resource Economics, UCB .

    Allee, T. & Peinhardt, C. (2011), Contingent credibility: The impact of investment treaty violationson foreign directinvestment, International Organization 65(3), 401432.

    Banks, A. & Wilson, K. A. (2005), Cross-national time-series data archive. jerusalem: Databanksinternational.

    Blonigen, B. A. & Piger, J. (2014), Determinants of foreign direct investment, Canadian Journal ofEconomics/Revue canadienne dconomique 47(3), 775812.

    BRASIL (2002), CMARA DOS DEPUTADOS. Ata da 234a Sesso Solene[lower house - minutes ofthe 234a solemn session]. Accessed: 04-23-2016.

    Brauch, M. D. (2015), The brazilmozambique and brazilangola cooperation and in-vestment facilitation agreements (cifas): A descriptive overview, URL: https://www.iisd. org/itn/2015/05/21/the-brazil-mozambique-andbrazil-angola-cooperation-and-investment-facilitation-agreements-cifas-adescriptive-overview (accessed: 07.07. 2015) .

    Bubb, R. J. & Rose-Ackerman, S. (2007), Bits and bargains: Strategic aspects of bilateral andmultilateral regulation of foreign investment, International Review of Law and Economics 27(3), 291311.

    Bthe, T. & Milner, H. V. (2009), Bilateral investment treaties and foreign direct investment: Apolitical analysis., in K. P. Sauvant & L. E. Sachs, eds, The effect of treaties on foreign directinvestment: Bilateral investment treaties, double taxation treaties, and investment flows, OxfordUniversity Press, Oxford, pp. 171224.

    Chinn, M. D. & Ito, H. (2008), A new measure of financial openness, Journal of comparative policyanalysis 10(3), 309322.

    Dolzer, R. (2005), The impact of international investment treaties on domestic administrative law,New York University Journal of International Law & Politics 37, 953969.

    Egger, P. & Pfaffermayr, M. (2004), The impact of bilateral investment treaties on foreign directinvestment, Journal of comparative economics 32(4), 788804.


  • Gabriel, V. (2016), The new brazilian cooperation and facilitation investment agreement: Ananalysis of the conflict resolution mechanism in light of the theory of the shadow of the law,Conflict Resolution Quarterly 34(2), 141161.

    Haftel, Y. Z. (2010), Ratification counts: Us investment treaties and fdi flows into developingcountries, Review of International Political Economy 17(2), 348377.

    Hallward-Driemeier, M. (2003), Do bilateral investment treaties attract fdi? only a bit. . . and theycould bite, World Bank Policy Research Working Paper No. 3121.

    Jakobsen, J. & De Soysa, I. (2006), Do foreign investors punish democracy? theory and empirics,Kyklos 59(3, 383410.

    Kalicki, J. & Medeiros, S. (2008), Investment arbitration in brazil, Arbitration International 24(3), 423446.

    Kerner, A. (2009), Why should i believe you? the costs and consequences of bilateral investmenttreaties, International Studies Quarterly 53(1), 73102.

    Laeven, L. & Valencia, F. (2008), Systemic banking crises: a new database, IMF working paperspp. 178.

    Lemos, L. B. & Campello, D. (2015), The non-ratification of bilateral investment treaties inbrazil: A story of conflict in a land of cooperation, Review of International Political Economy22(5), 10551086.

    Li, Q. & Resnick, A. (2003), Reversal of fortunes: Democratic institutions and foreign directinvestment inflows to developing countries., International organization 57(01), 175211.

    Mohr, L. B. (1995), Impact Analysis for Program Evaluation, 2nd Edition, Sage University Press,Thousand Oaks, CA.

    Morosini, F. & Ratton, M. (2015), The brazilian agreement on cooperation and facilitation ofinvestments (acfi): A new formula for international investment agreements?, Investment TreatyNews 6(3).

    Motta Veiga, P. (2004), Foreign direct investment in brazil: regulation, flows and contri-bution to development, International Institute for Sustainable Development, http://www. pdf 7(28), 05.

    Neumayer, E. & Spess, L. (2005), Do bilateral investment treaties increase foreign direct investmentto developing countries?, World Development 33(10), 15671585.

    Peinhardt, C. & Wellhausen, R. L. (2016), Withdrawing from investment treaties but protectinginvestment.

    Peterson, L. E., international du dveloppement durable, I. & du dveloppement et de la coopra-tion, S. D. (2004), Bilateral investment treaties and development policy-making, IISD.

    PRS-Group (2007), International country risk guide: Researchers data set. East Syracuse,N.Y.:PRS Group.

    Rose-Ackerman, S. & Tobin, J. (2005), Foreign direct investment and the business environmentin developing countries: The impact of bilateral investment treaties, Yale Law and EconomicsResearch Paper No. 293.


  • Salacuse, J. & Sullivan, N. (2005), Do bits really work? an evaluation of bilateral investmenttreaties and their grand bargain, Harvard International Law Journal 46(1), 4667.

    Sauvant, K. P. & Sachs, L. E. (2009), The effect of treaties on foreign direct investment: Bilateral investmenttreaties, double taxation treaties, and investment flows, Oxford University Press Oxford.

    UNCTAD (2001), World investment report 2001 - promoting linkages, UNCTAD .

    Walde, T. (2000), Treaties and regulatory risk in infrastructure investment, Journal of World Trade34(2), 161.

    Wellhausen, R. L. (2014), The shield of nationality: When governments break contracts with foreign firms,Cambridge University Press.

    Welsh, N. A., Schneider, A. K. & Rimpfel, K. (2014), Using the theories of exit, voice, loyalty, andprocedural justice to reconceptualize brazils rejection of bilateral investment treaties, Wash. UJL& Poly 45(1), 1509.

    World Bank (2004), Global development finance: Harnessing cyclical gains for development, TheWorld Bank .

    Yackee, J. W. (2007), Conceptual difficulties in the empirical study of bilateral investment treaties,Brook. J. Intl L. 33, 405.


    IntroductionBITs and FDIBrazil and BITs

    MethodsResultsPlacebo and Robustness ChecksDiscussionReferences