Governance, Globalization, and Selection into Foreign DirectInvestment∗

Koen Berden† Jeffrey H. Bergstrand‡ and Eva van Etten§

April 6, 2012

Abstract

Unlike the large literature on “democracy and trade,” there is a much smaller literature onthe effect of the level of democracy in a nation on the level of foreign direct investment (FDI).These few studies reveal mixed empirical results, and surprisingly only one study has examinedbilateral FDI flows. Moreover, few of these studies use multiple governance indicators separatingthe “pluralism” effect of democratic institutions from the “good governance” effect, there are nostudies on democratic institutions’ various effects on the level of FDI relative to trade, and thereare no studies of democratic institutions’ various effects on the selection of countries into FDI.We focus on three potential contributions. First, we examine the simultaneous effects of the (six)World Bank’s Worldwide Governance Indicators (WGIs) – which allow separating the effects ofpluralism from those of five other good governance measures – on bilateral trade, FDI, and FDIrelative to trade using state-of-the-art gravity specifications motivated by the general equilibriumKnowledge-and-Physical-Capital model in Bergstrand and Egger (2007, 2010, 2012). Second, wefind strong evidence that – after accounting for host governments’ effectiveness in various roles ofgood governance – a higher level of pluralism as measured by the WGIs’ Voice and Accountabilityindex reduces trade levels, likely by increasing the “voice” of more protectionist less-skilled workers,but not FDI levels. Moreover, we find qualitatively different effects of several other WGIs – such aspolitical stability and government effectiveness – on trade versus FDI flows. Third, we account forfirm heterogeneity alongside a large number of zeros in bilateral FDI flows using recent advancesin gravity modeling. We distinguish between the intensive and extensive margins and show thatpluralism (as measured by Voice and Accountability) affects FDI inflows negatively at the (country)intensive margin, but positively at the extensive margin.

Key words: Foreign direct investment; International trade; DemocracyJEL classification: F1; F2

∗Acknowledgements: Acknowledgements will be added later.†Affiliation: ECORYS and Erasmus University. Address: ECORYS, Watermanweg 44, 3067 GG Rotterdam, The

Netherlands. E-mail: Koen.Berden@ecorys.com.‡Affiliation: Department of Finance, Mendoza College of Business, and Kellogg Institute for International Stud-

ies, University of Notre Dame, and CESifo Munich. Address: Department of Finance, Mendoza College of Business,University of Notre Dame, Notre Dame, IN 46556 USA. E-mail: bergstrand.1@nd.edu.

§Affiliation: Erasmus University. Address: Campus Woudestein, Burgemeester Oudlaan 50, 3062 PA Rotterdam,The Netherlands. E-mail:

1 Introduction

“In each case, the increased pluralism ensured by democratic insti-

tutions generates policy outcomes that reduce the MNE’s degree of

freedom in the host developing country [and consequently FDI inflows].

On the other hand, democratic institutions promote FDI inflows by

strengthening property rights protection.” (Li and Resnick, 2003,

p. 177)

Two of the main channels for the economic globalization of the world economy – international

trade and foreign direct investment (FDI) flows – continue to advance at a rapid pace. First, while

much is known empirically about the determinants of both bilateral trade and bilateral FDI flows

separately, our knowledge about the empirical relationship between FDI and trade flows – either

cross-sectionally in a particular year or over time – is more limited. Most empirical studies have

investigated trade or FDI flows, not both simultaneously. Second, researchers have also typically

studied the impact of “democratic institutions” on FDI and trade separately. The results to date

are mixed; in some studies, indexes of democracy increase trade (FDI) and in others it reduces trade

(FDI). As the introductory quote above suggests, democratic institutions tend to foster “pluralism,”

which may have a negative effect on trade and FDI. However, democratic institutions also tend to

foster “property rights protection,” which may have a positive effect on trade and FDI.

No study has yet looked systematically at measures of “pluralism” alongside other aspects of

“good governance” on bilateral international trade and FDI flows simultaneously using state-of-the-

art gravity equations in a unified approach motivated by a formal general equilibrium model – much

less the influence of such factors on FDI relative to trade.1 Moreover, while researchers have just

started to examine the impact of governance indicators on selection of countries (and/or firms) into

trade, no study has yet examined the influence of pluralism alongside other governance indicators

on selection of countries (or firms) into FDI. This is the first study to our knowledge that examines

systematically how various measures of governance influence bilateral FDI flows relative to bilateral

trade flows and also the selection of countries into FDI (i.e., the country extensive FDI margin), also

accounting for firm heterogeneity (i.e., the firm margin).2

1Even the excellent recent study in this journal by Asiedu and Lien (2011) uses multilateral FDI stocks. The literaturewill be reviewed below.

2Eicher, Helfman, and Lenkoski (2011) have explored determinants of the margins of FDI using Bayesian ModelAveraging, based upon a gravity-equation approach. Also, Globerman and Shapiro (2003) appears to be the lone studyfirst examining selection into FDI, using bilateral U.S. FDI flows with numerous other countries (for three consecutiveyears). However, each of those authors’ probit estimations included only a constant, foreign GDP, and only one WorldwideGovernance Indicator (at a time); consequently, their specifications precluded separating the effects simultaneously of

2

First, international trade and FDI flows are likely influenced in reality simultaneously by common

factors. Trade does not cause FDI, nor does the reverse hold; firms select into or enhance their level

of trade or FDI based upon common economic and political factors. Not until theoretical research on

the general equilibrium determinants of foreign affiliate sales (FAS) and international trade flows in

Helpman (1984), Markusen (1984), Markusen and Venables (1998, 2000), and Markusen (2002) using

2-country, 2-factor, 2-good models did economists develop a more systematic framework for under-

standing conceptually the determinants of the levels of multinational firms’ foreign affiliate production

and sales (FAS) in foreign markets simultaneously with the levels of national firms’ production and

export decisions to various markets. Carr, Markusen, and Maskus (2001), Blonigen, Davies, and Head

(2003), and Markusen and Maskus (2001, 2002) used the 2x2x2 “Knowledge-Capital” model to moti-

vate empirical specifications for FAS and trade. More recently, Bergstrand and Egger (2007) extended

these 2x2x2 general equilibrium models of FAS and trade to include three countries – to explain the

influences of economic size and similarity, relative factor endowments, and investment and trade costs

on explaining the behavior of bilateral flows of trade, FDI, and FAS in a world with more than two

countries – and allowing for three factors of production – unskilled labor, skilled labor, and physical

capital – to help explain the complementarity of trade, FDI, and FAS flows found in aggregate data.

We draw upon these recent theoretical developments to help motivate estimating state-of-the-art grav-

ity equations of trade and FDI flows – the most common specification for explaining such flows – and

to explore empirically the pattern of FDI flows relative to trade flows.3 While rigorous theoretical

economic foundations for gravity models of trade have existed and evolved since Anderson (1979),

Bergstrand (1985), and Helpman and Krugman (1985), no such foundation has been formulated for

gravity equations of FDI and FAS flows until the 3-country, 3-factor, 2-good model in Bergstrand and

Egger (2007), even though the gravity model works extremely well for these flows also, cf. Blonigen

and Piger (2011) and Eicher, Helfman, and Lenkoski (2011).4

Second, although economists and political scientists have examined political determinants of trade

for decades, the literature on political determinants of FDI is much smaller, with a scant literature

to date examining the effect of democracy on FDI and finding conflicting empirical effects of various

measures of democracy. Moreover, only one study has used gravity equations with bilateral FDI flows

pluralism from the effects of other measures of good governance. Moreover, that study ignored accounting for firmheterogeneity.

3Since 2003, researchers have augmented “traditional” trade gravity equations with “multilateral price” (or resistance)terms for each country, to account for rest-of-world (ROW) effects, cf., Anderson and van Wincoop (2003, 2004). Inestimation, various methods have been used to account for endogeneity bias created by these terms. To date, alternativemethods to account for these terms in estimation include country-specific fixed effects, nonlinear structural estimation,or log-linear approximations of the underlying nonlinear price terms, cf., Baier and Bergstrand (2009) and Bergstrand,Egger, and Larch (2012) on these issues. We address these in detail later.

4See Bergstrand and Egger (2011b) for a survey of the gravity-equation literature regarding trade and FDI.

3

to focus on the effects of democratic institutions. We will review this sparse literature briefly later.

However, we note that Kaufmann, Kraay, and Mastruzzi (2007a) describe a panel data set of – what

many researchers consider – the “best existing measures of the quality of political institutions” (Kurtz

and Schrank, 2007, p. 539). The Kaufmann, Kraay, and Mastruzzi (2007a) “Worldwide Governance

Indicators” (WGIs), constructed under the auspices of the World Bank, aggregate numerous measures

of democracy and governance into six well defined groupings. Following Alcala and Ciccone (2004)

and Badinger (2008), we employ these six well known World Bank WGIs (explained in detail below) to

address the ambiguity surrounding the effects of governance indicators – and, in particular, measures of

“democracy” – on FDI and trade flows. For the impatient reader, we note that the ambiguous finding

in the literature relating democracy to FDI can be partially explained by the types of variables used

to capture democratic institutions. For instance, we find that measures associated with regulatory

quality faced by firms have positive effects on attracting FDI and trade flows, but indexes of political

stability and government effectiveness have different effects on FDI relative to trade. Moreover, a

measure of “Voice and Accountability” for citizens tends to have negative effects on trade inflows (but

can increase FDI relative to trade), likely associated with discrimination against foreign exporters in

favor of domestic firms by more protectionist members of society due to “pluralism.”

Third, while Voice and Accountability – the WGI index most closely associated with the democratic

notion of “pluralism” (i.e., freedom of speech, political participation, assembly, etc.) – is negatively

related to trade levels in our results, no one has yet examined how governance indexes influence the

selection of countries into FDI. This suggests the importance of examining separately the (country)

intensive versus the extensive margins of FDI, in the context of fixed exporting and investment costs

and firm heterogeneity.5 While theoretical developments regarding the roles of fixed exporting costs

and firm heterogeneity have spanned the trade literature in the last decade, the role of fixed FDI costs

and firm heterogeneity for explaining selection of countries into FDI flows is less developed with the

exception of a few papers, cf., Helpman, Melitz, and Yeaple (2004), Yeaple (2009), and Ramondo and

Rodriguez-Clare (2009). In the spirit of this new literature, we examine the potential impact of the

WGIs on the selection of countries into FDI. Notably, this is the first study to find that “Voice and

Accountability” in the host country – which tends to have a negative impact on their level of FDI

inflows – has a significant positive effect on the selection by home countries of FDI into hosts. This

suggests that – for developing countries searching for more FDI inflows – not only good governance in

the host country matters for FDI entry of new capital-exporting countries and firms, but also strong

5The intensive margin denotes the amount of trade (FDI) of existing firms already exporting (with plants abroad),while the extensive margin refers to the number of exporting (FDI exporting) firms, influencing whether or not a countryis exporting (investing). A zero trade (FDI) flow from one country to another is interpreted in this context as no firmsexporting (investing) from the origin country.

4

democratic institutions fostering pluralism also tend to increase the likelihood of selection of positive

FDI into a host country.

The remainder of the paper is as follows. In section 2, we summarize the analytical framework

behind our economic determinants of trade and FDI flows, the motivation for using the gravity equation

in the econometric work, and the role of democracy and the Worldwide Governance Indicators in our

analysis. Section 3 addresses the data and econometric issues. Section 4 presents the main empirical

results using ordinary least squares (OLS) and a “traditional” gravity equation framework and presents

some novel results explaining trade flows, FDI flows, and the ratio of FDI to trade. Section 5 provides a

sensitivity analysis of our results to account for bias attributable to heteroskedasticity, bias attributable

to ignoring the large numbers of zeros in FDI flows in the data set, and bias attributable to omitting

“multilateral price” (or resistance) terms. Section 6 accounts for potential biases introduced by country

sample selection and firm heterogeneity. Section 7 provides conclusions.

2 Framework and Hypotheses

The purpose of this study is to examine empirically: economic determinants of bilateral FDI, trade, and

FDI relative to trade flows; the influences of “voice and accountability” relative to other measures of

governance on FDI, trade, and FDI-relative-to-trade flows; and the influence of these determinants on

countries’ selection into positive bilateral FDI flows (accounting also for firm heterogeneity).6 As noted

earlier, while the literature on “democracy and trade” is quite large, the literature on “democracy and

FDI” is considerably smaller, and it is startling that only one empirical study has focused on explaining

bilateral FDI flows and used a gravity methodology. Moreover, no study has looked at the effects of

multiple governance indicators on bilateral FDI and trade simultaneously using a common state-of-

the-art gravity framework motivated by a formal general equilibrium model or selection into bilateral

FDI flows using the Helpman, Melitz, and Rubinstein (2008) methodology.

2.1 Motivating the Estimation Framework for Explaining Bilateral Trade and FDI

Flows

Most econometric studies that have examined determinants of bilateral trade flows and of bilateral FDI

flows have used the gravity equation, cf., Blonigen (2005).7 In this paper, we use the gravity equation as

well. As Blonigen and Piger (2011) recently noted, Carr, Markusen and Maskus (2001) and Bergstrand

6Even though our data set includes both FDI and trade flows for the same pairings of 28 source countries with124 destination countries, our sample of bilateral trade flows is composed of positive flows only, precluding examiningselection into trade.

7Empirically, it is standard to investigate bilateral FDI “stocks.” Consistent with the literature, we use FDI stockdata (not flows per se), cf., Blonigen and Piger (2011).

5

and Egger (2007) provided theoretical general equilibrium models of multinational enterprises’ and

national exporting firms’ behavior to suggest economic factors that explain simultaneously trade and

foreign affiliate sales (FAS). Carr, Markusen, and Maskus (2001) and Markusen and Maskus (2001,

2002) use the 2-good, 2-factor, 2-country “Knowledge Capital” model of Markusen (2002), which is

based upon earlier work in Helpman (1984), Markusen (1984), Markusen and Venables (1998, 2000),

and several other papers of Markusen with various coauthors. This framework provided foundations

for the roles of two countries’ GDP sizes, GDP similarities, relative skilled-to-unskilled-labor shares,

and bilateral trade and FAS costs in influencing the levels of bilateral FAS.8

Bergstrand and Egger (2007) is an extension of the 2x2x2 “Knowledge-Capital” (KC) model in

Markusen (2002), also with national exporters (NEs), horizontal multinational enterprises (MNEs),

and vertical MNEs. Prior to Bergstrand and Egger (2007), the limitation of the general equilibrium

KC model to two factors (skilled and unskilled labor) and two countries did not allow a theoretical

foundation for the observed complementarity of bilateral trade and FDI flows for identical countries

(once all other factors known to influence trade and FDI were accounted for), nor did it generate a

theoretical foundation for using the “gravity equation” for explaining empirically FDI flows – much

less explaining simultaneously the use of gravity equations for both FDI and trade flows.

Motivated by the puzzle in the Markusen-Venables 2x2x2 general equilibrium model of NEs and

MNEs that two countries with identical relative factor endowments maximize their bilateral foreign

affiliate sales when their absolute factor endowments (and hence GDPs) are identical but have zero

bilateral trade – which is empirically rejected – Bergstrand and Egger (2007) introduced a third fac-

tor (physical capital) to resolve this puzzle. Then, they introduced a third country (Rest-of-World,

or ROW ) to readily motivate theoretically gravity equations of bilateral FDI and trade simultane-

ously. Although introducing a third factor implies coexistence of NEs and HMNEs for identically-sized

economies, this extension cannot explain empirically the ”complementarity” of bilateral trade and FDI

flows to GDP similarity. Typical empirical gravity equations of international trade and FDI tend to

suggest that both trade and FDI from country i to country j should be positively related to the size

and similarity of their GDPs. However, the introduction of a third country – ROW – to the three-

factor “Knowledge-and-Physical-Capital” (KAPC) model can explain readily the complementarity of

bilateral trade, FDI, and FAS to changes in a pair of countries’ economic size and similarity as typ-

ical to gravity equations. In a two-country world, gross multilateral and bilateral trade (or foreign

affiliate sales) are identical; NEs and HMNEs must substitute for one another when the two countries

are identically sized in the face of trade and investment costs. However, introducing a third country

(along with imperfect mobility of the services of physical capital) allows two countries’ trade, FDI, and

8Moreover, as Markusen (2002) notes, this 2x2x2 framework cannot explain FDI flows, only FAS flows.

6

foreign affiliate sales (FAS) to co-vary positively with increases in these two countries’ GDP similarity

because the “substitution effect” associated with exogenous trade-to-investment costs is potentially

offset by a “complementarity effect” generated by endogenous relative prices of physical-to-human

capital interacting with the three countries’ economic sizes. With three countries, both bilateral trade

and FAS are maximized when a pair of countries’ GDPs are identical, unlike a two-country world.

Moreover, the presence of the third country can explain why FDI from one country to another is not

maximized when GDPs are perfectly identical – which is actually observed empirically.9

Given the potentially large number of variables that have been used in the extensive empirical

literature using gravity equations to explain bilateral FDI stocks, Blonigen and Piger (2011) recently

conducted a Bayesian Moving Average analysis to allow one to select (for the gravity equation of

FDI) from an enormous set of candidate variables the ones most likely to explain FDI stocks. The

variables most likely to explain (the log of) FDI bilateral stocks – using a cutoff threshold of 100

percent – included home and host countries’ (log) real GDPs, (log) bilateral distance, the (log of)

home country’s per capita real GDP – which are all standard gravity equation variables – and relative

skilled labor endowments (specifically, the squared difference in the two countries’ shares of unskilled

labor). The variables with likelihoods ranging from 90-100 percent included common official language,

host country’s remoteness, home country capital-labor ratio, host country’s urban concentration ratio,

and regional trade agreement dummy – many of which are often included in gravity equations.10

Although the remaining variables examined all had likelihoods below 80 percent, it is worth noting

for our study at hand the variables that still had material likelihood of inclusion, in particular, those

with probabilities of 20-80 percent. These variables were customs union dummy, former colonial

relationship, host country’s trade openness, host country’s skill level, host country’s corporate tax

rate, host country’s per capita real GDP, and host country’s political rights. Thus, measures of

democracy and other governance indicators surface as factors potentially important in explaining FDI

flows.

Consequently, given the theoretical motivation in Bergstrand and Egger (2007) and the econometric

motivation from the Bayesian Moving Average (BMA) analysis in Blonigen and Piger (2011), our

empirical specification for the gravity equation includes most of the explanatory variables suggested

in Blonigen and Piger (2011) as having systematic material influence in the BMA analysis (above 20

percent threshold): source and destination countries’ GDPs and per capita GDPs, bilateral distance,

9One of the important theoretical results in Bergstrand and Egger (2007) was that – in the context of their numericalgeneral equilibrium model – for FDI flows the home country’s GDP elasticity should exceed the host country’s GDPelasticity, whereas for trade flows the exporter’s and importer’s GDP elasticities should be virtually identical. SeeBergstrand and Egger (2007) for details for these conclusions. We address this issue here as well.

10In many cases, regional trade agreements include bilateral FDI liberalization provisions also.

7

common language dummy, common colonial relationship dummy, and destination country’s political

variables. At this time, we exclude free trade agreement and customs union dummies; there are

numerous econometric (endogeneity) issues associated with such dummies in gravity equations, which

have been addressed elsewhere (cf., Baier and Bergstrand, 2007) and are beyond the scope of this paper.

We also for now exclude source country’s physical capital-labor ratio and country-pairs’ differences in

unskilled labor shares, which have been addressed elsewhere (cf., Bergstrand and Egger, 2011a) and

are beyond the scope of this paper. Trade “openness” is omitted for reasons discussed earlier; given

our theoretical context, trade and FDI flows are determined simultaneously by common factors.11

In the next section, we examine the well-known World Bank Worldwide Governance Indicators

(used in several previous influential economic analyses) as potential determinants of bilateral FDI,

trade, and FDI relative to trade, as well as selection into FDI.

2.2 The Worldwide Governance Indicators

The title of our paper, ”Governance, Globalization, and Selection into FDI,” reveals our emphasis on

examining the effects of governance – not just “democracy” per se – on FDI, trade, FDI relative to

trade, and selection into FDI. Our approach suggests that different elements of governance can have

complementary or conflicting effects on two of the three main channels of economic globalization: FDI

and trade.12 Our approach is primarily an empirical one, to let the data speak about how various

measures of governance affect FDI and trade. However, revealed correlations will be examined to see

if they have economically feasible interpretations.

Milner and Mukherjee (2009) provide one of the most comprehensive recent literature reviews of

the interaction between “democracy” and “trade.”13 To date, most empirical work analyzing the

interactions of these variables has examined causality running from democracy to international trade.

Milner and Mukherjee (2009) summarize the literature as showing that increased democracy tends to

increase the trade openness of countries significantly. However, when examining the literature report-

ing surveys of workers “trade preferences,” the literature tends to suggest that low-skilled/unskilled

workers in countries – either developed or developing – tend to be against trade openness. The authors

note several surveys that confirm this view, cf., Scheve and Slaughter (2001) and Mayda and Rodrik

(2005). Milner and Mukherjee (2009) conclude the widening of a “skill premium“ bias in countries fol-

11For now, we also exclude the destination country’s urban concentration ratio and corporate tax rate; we leave theirinclusion for future research.

12The third main channel is migration, not addressed here.13Milner and Mukherjee (2009) examine both the relationships between democracy with trade as well as between

democracy and “capital flows” (or capital account openness). Since their survey is recent and comprehensive, we referthe reader to that paper for a full list of the important contributions. Since they do not address the literature on therelationship between democracy and FDI, this survey does not shed light on the latter. However, a useful review of thesparse literature on democracy and FDI is provided recently in Asiedu and Lien (2011), discussed shortly.

8

lowing trade liberalization may affect preferences, attitudes, and the “voice” of workers against trade

liberalization, even though broad measures of “democracy” may be positively correlated with trade.

Also, Milner and Mukherjee (2009) summarize that the evidence to date on the reverse direction of

causality – from trade to democracy – is weak, at best.

By contrast, the literature examining the effect of democracy on FDI is much smaller. Recently,

Asiedu and Lien (2011) noted that within the context of a vast empirical literature on the determinants

of FDI “only a few of the studies include democracy as an explanatory variable” (p. 101). They noted

only 12 published studies that have included democracy as a determinant of FDI, and only two were

published before 2000. Of these 12 studies, eight found significant positive effects of democracy on

FDI, three found no significant effects, and only one found a significantly negative effect of democracy

on FDI.14 Of the eight studies finding positive relationships, several of the studies were limited by

numerous aspects: (1) ad hoc FDI specifications lacking rigorous theoretical foundations; (2) including

only measures of democracy per se but excluding representation of measures of property rights or

“good governance”; (3) small samples typically using multilateral FDI inflows by country, rather than

bilateral flows; and/or (4) inclusion in FDI regressions of measures of trade (to reflect “openness”), cf.,

Harms and Ursprung (2002), Jensen (2003), Jakobsen (2006), Adam and Filippaios (2007), Busse and

Hefeker (2007). None of these 12 studies used bilateral FDI flows or a gravity-equation methodology.15

Asiedu and Lien (2011) note that only one published study – Li and Resnick (2003) – found a

significant negative effect of democracy on FDI. The notable distinction of Li and Resnick (2003) was

the additional inclusion of numerous measures of “property rights” on the RHS. With those variables

included, democracy per se had a negative impact. However, in response to Li and Resnick (2003),

Jakobsen and de Soysa (2006) showed that – if one doubles the sample size from 50 countries to nearly

100 and uses the logarithm of multilateral FDI flows (rather than the level) – this finding is reversed;

democracy and property rights indexes have complementary positive effects on FDI.

Li and Reuveny (2003) and Li and Resnick (2003) are straightforward regarding the potentially

14They noted that Rodrik (1996), Harms and Ursprung (2002), Jensen (2003), Busse (2004), Jakobsen (2006), Jakobsenand de Souya (2006), Adam and Filippaios (2007) and Busse and Hefeker (2007) found positive effects; Oneal (1994),Alesina and Dollar (2000) and Buthe and Milner (2008) found no significant effects; and Li and Resnick (2003) found asignificant negative effect.

15Globerman and Shapiro (2002) examined the influences of the Worldwide Governance Indicators separately on FDI,but also used multilateral FDI levels and did not include simultaneously all the Worldwide Governance Indicators. Asnoted earlier, Globerman and Shapiro (2003) appears to be the lone study first examining selection into FDI, usingbilateral U.S. FDI flows with numerous other countries (for three consecutive years), and then accounting for selectionbias in subsequent regressions explaining FDI levels. However, each of those authors’ probit and second stage estimationsincluded only a constant, foreign GDP, and only one Worldwide Governance Indicator (at a time); consequently, theirspecifications precluded separating the effects simultaneously of pluralism from the effects of other measures of goodgovernance. Moreover, that study ignored accounting for firm heterogeneity. The lone study we have found using agravity methodology and bilateral FDI stocks is Benassy-Quere, Coupet, and Mayer (2007). Of course, Blonigen andPiger (2011) and Eicher, Helfman, and Lenkoski (2011) included democracy measures in their respective BMA analyses,but did not focus on such variables.

9

conflicting effects of “democracy” on FDI, noting that democratic “institutions” (such as regulatory

quality) tend to strengthen property rights’ protections, thus enhancing FDI. However, democratic

“constraints” (such as voice and accountability for citizens) tend to weaken market powers of MNEs,

diminishing FDI. Thus, indexes of higher pluralism – such as “voice of citizens” along with “account-

ability of government to citizens” – may increase unskilled labor’s voice and be correlated negatively

with trade and FDI – even though the more transparent, less corrupt, and more effective governance

associated with democracies may well lead to more trade and FDI.16

The mixed empirical outcomes in the empirical literature on democracy and globalization suggests

the need for a study using a broader set of measures of “governance” – of which variables related to

pluralism per se can be isolated – to separate the potentially conflicting effects of pluralism from other

governance structures. In this paper, we employ the well known “Worldwide Governance Indicators.”

The WGIs are very useful because they include six indicators that span a wide array of factors that

can potentially affect FDI, trade, and even FDI relative to trade. Important studies such as Alcala

and Ciccone (2004) and Badinger (2008) employed the WGIs, constructed under the auspices of the

World Bank, because they are considered in a recent survey “probably the most carefully constructed

governance indicators” (Arndt and Oman, 2006). While there is now an emerging literature on these

indicators, two of the notable features of them is that by aggregating over numerous sources they

reduce dramatically measurement error and they are now constructed annually over a much larger

number of countries than most other governance indicators.17 While critiques of these indicators have

been made because of their prominent adoption, they are widely respected indicators and the most

suitable indicators for this study, cf., Kaufmann, Kraay, and Mastruzzi (2006, 2007a, 2007b, 2007c),

Kurtz and Schrank (2007), Arndt and Oman (2006) and references therein; we refer the reader to this

literature for more detail.

We now list what each of the six indicators measures, as summarized in Kaufmann. Kraay, and

Mastruzzi (2007a):

1.) Voice and Accountability (VA): This variable measures the extent to which a country’s citizens

are able to participate in selecting their government, as well as freedom of expression, of association,

and of media. Of the six WGIs, this variable best captures most individuals’ notion of how a democratic

16Many of the studies in the political science journals cited in the previous footnote articulate the rationale for apotential negative effect of democracy on FDI. Li and Resnick (2003) are particularly clear that the increased pluralismassociated with democracies tends to weaken MNEs’ freedom in host countries, tending to reduce FDI inflows. Theynote that democratic constraints over elected politicians tend to weaken the oligopolistic or monopolistic positions ofMNEs, democratic constraints bind host governments from offering generous financial incentives for FDI, and broadaccess to elected officials and wide political participation offer institutionalized routes through which businesses can seekprotection.

17The indicators are constructed annually beginning in 2002; prior to 2002, the indicators were constructed for 1996,1998, and 2000. They cover now over 200 countries and use information from 31 different data sources from 25 differentorganizations, including some of the most prominent political science indicators of democracy.

10

institution fostering voice and accountability affects pluralism.18 This variable should be negatively

related to trade and FDI.

2.) Political Stability (PS ): This variable measures perceptions of the likelihood that the govern-

ment will not be destabilized or overthrown by unconstitutional or violent means. For FDI, MNEs

should tend to prefer a stable to an unstable host government, due to risk of expropriation. For trade,

political stability need not be a crucial determinant because of the absence of risk of expropriation of

plant and equipment.

3.) Government Effectiveness (GE ): This variable measures the quality of public services, of the

civil service (and its degree of independence), of policy formation process and implementation, and of

the government’s commitment to implementing policies. For FDI and trade, one would expect that

foreign companies would prefer an effective host country government.

4.) Regulatory Quality (RQ): This variable measures the ability of the government to formulate

and implement sound policies and regulations that permit and promote private sector development.

Of the six indicators, this one should be very important for enhancing both FDI and trade.

5.) Rule of Law (RL): This variable measures the extent to which agents have confidence in and

abide by the rules of society, and in particular the quality of contract enforcement, the police, and the

courts. This should be important for both FDI and trade.

6.) Control of Corruption (CC ): This variable measures the extent to which public power is not

exercised for private gain, including both petty and grand forms of corruption, as well as “capture”

of the state by elites and private interests. This could be important both for FDI and trade.

As discussed earlier, the theoretical literature has suggested that a measure of greater pluralism –

such as Voice and Accountability – can tend to lower imports of goods as well as decrease host country

FDI inflows. By contrast, the latter five governance indicators should tend to have a positive effect

on trade and/or FDI, as they should tend to reduce the transaction costs for an MNE to invest in a

host country or for an exporting firm to trade with an importing country.

There is – to our knowledge – no established literature on how any of these six factors would

influence the level of FDI relative to trade, much less selection into FDI. However, we will find that

these governance indicators influence trade and FDI differently – and thus affect FDI relative to trade

– and affect positive levels of FDI differently from selection into FDI.

18The frequently considered minimum requirements for a democracy are: regular, free and fair elections; divisions ofpowers; checks and balances; and inalienable human freedoms, rights, and voice.

11

3 Econometric and Data Issues

In the first part of this section, we discuss specification issues for the econometric work. In the second

part, we discuss the sources of data (other than the World Bank WGIs just described) used for the

empirical work.

3.1 Econometric Issues

In our main results, we will use a traditional specification for gravity equations, employ ordinary least

squares, and use only positive trade and FDI flows (i.e., we omit zeros). In later results, we will

consider alternative specifications motivated by either theoretical considerations, econometric ones, or

data ones (e.g., include zeros).

The first specification we use for (the logarithm of) bilateral trade, bilateral FDI, and the ratio of

bilateral FDI to bilateral trade is:

lnXijt =β0 + β1 lnGDPit + β2 lnGDPjt + β3 lnPCGDPit + β4 lnPCGDPjt + β5 lnDISTij

+ β6ADJij + β7LANGij + β8COLONYij + β9V Ajt + β10PSjt + β11GEjt + β12RQjt

+ β13RLjt + β14CCjt + ϵijt

(1)

We use three different specifications, employing different LHS variables for lnXijt. In column (3) of

Table 1, we use lnTRADEijt, which is the natural logarithm of the flow of merchandise trade from

exporting country i to importing country j (in year t). In column (4), we use lnFDIijt, which is the

natural log of the stock of foreign direct investment of home country i in host country j. FDI stocks

rather than flows are the standard in the empirical gravity equation literature on FDI “inflows,” cf.,

Blonigen (2005). Finally, in column (5), we use lnFDITRADEijt, which is the natural log of the

ratio of the FDI flow to the trade flow. In the main specifications, we only use positive values of FDI

and trade flows. In our data set, all trade flows have positive values but numerous FDI flows are zeros;

we address the zeros in FDI flows later. Data sources and countries included will be discussed later.

In equation (1), the following variables are included on the RHS. Variable lnGDPit (lnGDPjt) de-

notes the natural logarithm of the GDPs of country i (j) in year t. Variable lnPCGDPit (lnPCGDPjt)

denotes the logarithm of per capita GDP of country i (j) in year t. As the trade and FDI gravity

equation literature remains ambiguous about the interpretation of per capita GDPs in the gravity

equation, we remain agnostic and do not offer any firm prediction.19 Variable lnDISTij denotes the

19There has been no theoretical rationale for including per capita GDPs of home and host countries in the FDI gravityequation literature. In the trade gravity equation literature, Bergstrand (1989, 1990) remain the only studies that haveoffered a theoretical rationale for exporter and importer per capita GDPs. In those studies, exporter per capita GDPcan be interpreted in his context as a proxy for the capital-labor endowment ratio of country i; a positive (negative)

12

logarithm of the bilateral distance between the economic centers of countries i and j. ADJij is a

dummy variable assuming the value 1 (0) if countries i and j share (do not share) a common land

border. LANGij is a dummy variable assuming the value 1 (0) if countries i and j share (do not share)

a common language. COLONYij is a dummy variable assuming the value 1 if one of the countries

was a former colony of the other country, and 0 otherwise. Such variables have become standard

regressors in typical gravity equations as noted above in our discussion of Blonigen and Piger (2011).

All WGI indexes are included in level (not log-level) form; consequently, their coefficient estimates

will be considerably smaller (in absolute value terms) than the coefficients on variables in log form.

We consider in our robustness analysis later three modifications to the use of the standard gravity

equation used initially. The first modification is that – with OLS and several variables measured using

logarithms in the typical gravity equation – observations with zeros in the aggregate bilateral FDI

stocks are necessarily dropped and the coefficient estimates are potentially biased because of Jensen’s

inequality causing potential heteroskedasticity. Following Silva Santos and Tenreyro (2006) we also

run the specifications above accounting for heteroskedasticity using Poisson Pseudo-Maximum likeli-

hood (PPML) and then also including zeros.20 Second, as discussed extensively in Anderson (1979),

Bergstrand (1985), Anderson and van Wincoop (2003), Baier and Bergstrand (2009) and elsewhere,

the typical gravity equation is mis-specified when measures of “multilateral resistance” are omitted.

Because such multilateral resistance terms are country specific and time varying – like the World-

wide Governance Indicators – one cannot simply include exporter-and-time and importer-and-time

fixed effects, as this would eliminate the WGIs (i.e., perfect collinearity). Consequently, in the second

robustness specification, we use time-invariant exporter and importer fixed effects to account for all

time-invariant country specific influences (including the time-invariant portion of GDPs, per capita

GDPs, multilateral resistance, and the governance indicators); this specification also includes year

dummies to account for trend growth in world (FDI or trade) flows. Since the time variation in our

particular data set using the WGIs is very limited (we have only five cross-sections), this sensitivity

analysis will expectedly reduce considerably the influences of all the time-varying variables; neverthe-

less, we think it important to show these results. Third, because of the limited time variation in our

RHS variables of interest, in a third robustness analysis we consider a procedure suggested recently in

Baier and Bergstrand (2009) to account simultaneously for the multilateral resistance terms as well

as the WGIs. Baier and Bergstrand (2009) used a first-order log-linear Taylor-series approximation

coefficient estimate implies that the trade flows embody on average capital (labor) intensive goods. In that study, apositive (negative) coefficient estimate on the importer’s per capita GDP implies that the bundle of (aggregate) tradeflow from i to j is comprised on average of luxuries (necessities).

20As noted in Santos Silva and Tenreyro (2006) and the subsequent literature, there are two reasons for using PPML,the existence of zeros and heteroskedasticity-induced potential bias. Hence, even for our trade data set with no zeros,PPML will yield different coefficient estimates for the trade gravity equation than OLS.

13

method to isolate exogenous components of the multilateral price terms, which allows time- and cross-

section variation in the WGIs.21 Finally, to examine selection into FDI, we employ probit regressions

later to predict the probability of positive FDI flows and then implement the Helpman, Melitz, Ru-

binstein (2008) two-stage methodology to control for country-selection bias and for firm-heterogeneity

bias.

3.2 Data

The trade and FDI flow data are annual observations for the period 1997-2004. Cross-section time-

series bilateral FDI data are more difficult to obtain than bilateral trade flow data. The FDI inflows

are from the Organization for Economic Cooperation and Development (OECD) data base on Interna-

tional Direct Investment with 28 OECD countries as source countries and (potentially) 124 destination

countries. The bilateral trade flow data used are from the UN COMTRADE database.22

GDP and population data are from the International Monetary Fund’s International Financial

Statistics. Unfortunately, within the eight-year time frame of 1997-2004, the WGIs only exist for the

years 1998, 2000, and 2002-2004, as discussed in Kaufmann, Kraay, and Mastruzzi (2007a). This

limits our time dimension to only five years.23 Our WGI variables range between 0 and 100. The bi-

lateral distance (between economic centers) variable and adjacency, language and colonial relationship

dummies (defined earlier) were obtained from the CEPII website.

4 OLS Empirical Results

Table 1 presents the baseline empirical results using OLS and positive values of bilateral trade and

FDI flows. The table is organized as follows.24 Column (1) lists the RHS variables used in the main

empirical specification (equation (1)) which is the traditional gravity equation estimated using OLS.

Column (2) lists the expected coefficient signs when possible for specifications listed in columns (3)

and (4) only ; column (3) lists the results for the trade regression and column (4) lists those for the

FDI regression. Column (5) lists the results for the specification for FDI relative to trade. Hence,

21Baier and Bergstrand (2009) employ a first-order log-linear Taylor series expansion of the multilateral price terms inAnderson and van Wincoop (2003) to show how one can account for these price terms in estimation and in comparativestatics, without losing the informational content of other country-specific time-varying exogenous variables.

22This implies 17,220 potential gross flows for the five years. Both the trade flow data and the FDI flow data areconstrained by numerous missing observations. This leaves 12,229 trade flows observations (with no zeros) and 9,360FDI flow observations (and 4,840 positive FDI observations).

23Since years 1997, 1999, and 2001 were missing, we also tried interpolated values for the WGIs, generating three moreyears of observations. The results for all our specifications were not materially different using 5 or 8 years. We reportonly the results using the 5 years discussed.

24We use clustered standard errors, clustered around the destination-country variables to avoid overestimating thesignificance of the destination-country WGI variables’ coefficient estimates. Using “robust” standard errors often yieldedlower standard errors.

14

column (2)’s expected signs do not apply to the last specification in column (5); we will discuss later

expected signs for the specification for FDI relative to trade. In the case of per capita GDPs and

V Ajt, even for columns (3) and (4) the expected coefficient signs are unknown a priori.25 We discuss

the results for each of the three specifications in columns (3)-(5) in order.

Columns (3) and (4) provide the results for the typical gravity equations for the logarithm of

bilateral trade from exporter i to importer j and for the logarithm of the FDI from home country i to

host country j, respectively. Recall that, in our data set, all trade flows are positive, but only about

half of our FDI flows are positive; Table 1’s results uses only positive values of FDI (n = 4,840 in

Column (3)).

We organize our discussion below of variables’ coefficient estimates into three parts: (i) GDPs and

per capita GDPs; (ii) bilateral natural trade and investment cost variables (lnDIST,ADJ,LANG,

and COLONY ); and (iii) the WGI variables. First, consistent with Bergstrand and Egger’s (2007)

Knowledge-and-Physical-Capital model’s predictions, exporter and importer GDPs in column (3) and

home and host country GDPs in column (4) have positive effects on trade and FDI flows, respectively.

In fact, the larger GDP coefficient estimate for the home country relative to the host country in

column (4) is consistent with predicted coefficients in Bergstrand and Egger (2007). Regarding FDI

relative to trade in column (5), we note a statistically significant effect of country i’s GDP on FDI

relative to trade, and the economic effect is non-trivial, consistent with Bergstrand and Egger (2007).

Country i’s per capita GDP also has an economically and statistically significant effect on country i’s

trade outflow and FDI outflow to country j; this is consistent with evidence in Blonigen and Piger

(2011). It is also important to note that country j’s GDP has no perceptible impact on j’s bilateral

FDI inflow. This will be important later for econometric “identification” purposes when we isolate

intensive and extensive margin effects using the Helpman, Melitz, and Rubinstein (2008) methodology

to account for selection bias and firm heterogeneity. Moreover, i’s per capita GDP has an economically

and statistically significant effect on FDI relative to trade (from i to j), consistent with much firm-

level empirical results that multinational enterprises (MNEs) tend to be much more human-capital

and physical capital intensive in production than national exporting firms (NEs) and that developed

countries – which are relatively abundant in human and physical capital – tend to headquarter more

MNEs relative to NEs.

Second, consider the bilateral natural trade-and-FDI cost variables. First, (the log of) bilateral

distance, lnDISTij , has economically and statistically significant negative effects on trade and FDI

flows. The larger (in absolute terms) coefficient estimate in the trade-flow specification relative to the

25However, Blonigen and Piger (2011) suggest strong ex post empirical evidence that exporter (home country) percapita GDP is positively related to export (FDI) flows to an importer (host country).

15

FDI-flow specification – contributing to the large and positive coefficient estimate for FDI relative

to trade in column (5) – is typical. Distance is commonly considered to raise natural “trade costs”

of exporting from i to j. The smaller (absolute) coefficient for FDI can be readily explained by the

tension between horizontal FDI versus vertical FDI. Horizontal FDI is related to market access, and a

motivation for it is trade-cost “jumping.” Hence, the larger the bilateral distance between a country

pair, the larger their horizontal FDI (and associated MNE activity). However, some FDI is vertically

motivated, with headquarters country i investing in relatively low cost country j, goods are produced

in j, and then there are exports from j back to i as well as to the rest-of-world (ROW ). In this

case, larger bilateral distance raises trade costs, and tends to reduce vertical FDI from i to j. On

net, coefficient estimates for distance tend to be smaller for FDI relative to trade (in absolute terms),

as found here. Second, having a common land border (ADJij), common language (LANGij), and

common colonial history (COLONYij) tend to increase trade and FDI, as found in earlier studies. A

novel finding here is that all three variables tend to increase FDI relative to trade by economically

and statistically significant amounts.26

Third, consider the WGI variables’ effects. A prominent finding in Table 1 is that Voice and

Accountability (V Ajt) has a negative and statistically significant effect on trade and a negative but

insignificant effect on FDI, as shown in columns (3) and (4), respectively.27 Also, V Ajt has a statis-

tically insignificant positive effect on FDI relative to trade. In economic terms (with WGIs varying

between 0 and 100), a one-unit change in V Ajt decreases trade (FDI) by 0.012 (0.007) percent. V Ajt

is the WGI variable that most closely captures the degree of “pluralism” in a country. Our empirical

results are consistent with the Li and Resnick (2003) findings that more democratic pluralism in a

host country tends to lower trade and FDI inflows.

Also interesting, however, is the variance across the other governance indicators in terms of their

qualitative impacts on trade and FDI. Political Stability in the destination country (PSjt) tends to have

no effect on trade flows into j but a statistically significant impact on FDI into j. This empirical result

makes economic sense because FDI in a host country typically involves significant fixed investments

in plant and equipment, such that political instability in the host country (and risk of expropriation)

causes non-trivial investment costs for investors. By contrast, Government Effectiveness (GEjt) has

a statistically significant positive effect on trade, but no significant effect on FDI. Since this variable

measures the quality of public services, one might expect this to have a significant impact on both

trade and FDI, and while it does have a positive impact on both, only for trade is the coefficient

26This finding is novel because, as stated earlier, few studies examine empirically trade and FDI determinants simul-taneously.

27We note now that these two findings for V Ajt will tend to be robust across specifications throughout this study.

16

estimate statistically significantly different from zero. By contrast to PSjt and GEjt, Regulatory

Quality (RQjt) has a positive and statistically significant effect on both trade and FDI. However, the

effect of high regulatory quality has a disproportionately larger impact on FDI relative to trade. The

only WGI variable that has an unexpected negative coefficient estimate sign is Rule of Law (RLjt),

and it is statistically significant for FDI. However, this negative coefficient estimate for FDI flows for

RLjt will become insignificant later under more econometrically appropriate specifications. Moreover,

RLjt is highly correlated with several other WGIs, with the exception of Voice and Accountability.

Indeed, in separate regressions including each of the non-V Ajt WGI variables separately, all non-V Ajt

WGI variables have statistically significant positive effects on trade and FDI flows, including RLjt.

Finally, Control of Corruption (CCjt) has no perceptible effect on trade and FDI flows, when the

other WGI variables are included; when included in isolation from the other WGI variables CCjt has

a positive and statistically significant effect.

Overall, the fits of the equations are good. For the trade flow equation in column (3), the overall

R2 value is 83 percent, which is in line with previous trade gravity equations. For the FDI equation

in column (4), the overall R2 value is 66 percent. Typically, FDI gravity equations’ R2 values tend to

be less than those for trade flows. We will find later in our sensitivity analysis that both R2 values

can be improved by addressing some econometric concerns, and in fact we can also close the gap in

R2 values between the trade and FDI specifications. Moreover, because it is rare to have trade and

FDI flows examined simultaneously, as here, there is likely no previous estimate for FDI relative to

trade. Consequently, a novel finding is that we can explain 41 percent of the variation in the ratio of

FDI to trade.

5 Sensitivity Analysis

The gravity equation has developed a central role in the international trade literature, cf., recent

surveys by Anderson (2011) and Bergstrand and Egger (2011b), as well as a study by Arkolakis,

Costinot, and Rodriguez-Clare (2012). In the past ten years, theoretical advances and empirical

evidence have led to systematic modifications in empirical specifications of the trade gravity equation.

First, one of the notable advances raised in Santos Silva and Tenreyro (2006) is that heteroskedasticity

(attributable to Jensen’s inequality) and the large number of zeros in aggregate trade flows both

give rise to potential specification bias in log-linear gravity equations estimated using OLS. These

authors have argued in favor of estimating gravity equations using levels (rather than log-levels)

of variables using Poisson Pseudo Maximum Likelihood (PPML). In section 5.1, we consider this

alternative approach. Second, another notable advance raised systematically in Anderson and van

17

Wincoop (2003) is that proper estimation of bilateral trade gravity equations should account for third

countries’ prices. While the gravity equation literature has longed recognized (since the 1960s) that

bilateral trade flows are influenced by two countries’ GDPs relative to world GDP, the early gravity

equation literature was less receptive to including price terms, cf., Linnemann (1966) and Sapir (1978)

for relevant summaries. While Anderson (1979) and Bergstrand (1985) provided early formal general

equilibrium foundations for the gravity equation of trade that illustrated the potential importance

of recognizing “multilateral prices,” only since Eaton and Kortum (2002) and Anderson and van

Wincoop (2003) have empirical researchers addressed more systematically this issue. In section 5.2,

we account for endogeneity bias in our PPML estimates by introducing exporter and importer fixed

effects, a standard technique. Unfortunately, our WGI variables data set is constrained to only five

annual cross-sections. This leaves little time-series variation to capture the effects of trade and FDI

determinants. However, Baier and Bergstrand (2009) recently demonstrated using a first-order Taylor-

series “approximation method” that the multilateral resistance terms could be approximated using

exogenous factors, generating potentially unbiased coefficient estimates in gravity equations without

using fixed effects. In section 5.3, we account for potential endogeneity bias in the PPML estimates

using instead this recent ”approximation method” for multilateral prices.

5.1 PPML Estimates

Santos Silva and Tenreyro (2006) showed empirically and using a Monte Carlo analysis that one of

the important implications of Jensen’s inequality is that the standard practice of interpreting esti-

mates of coefficients from log-linear gravity models as elasticities can be misleading in the presence

of heteroskedasticity. That study proposed using a Poisson Pseudo-Maximum Likelihood (PPML)

estimator of the gravity model in level form (i.e., multiplicative in levels) to address the heteroskedas-

ticity that causes biased estimates using OLS on the log-linear form of gravity, such as we employed

in equation (1). Moreover, estimating the multiplicative (in levels) gravity equation allows inclusion

of flows having values of zero. Typically, zeros in trade and FDI flows have been dealt with using the

log-linear gravity equation by simply omitting them. Yet as Helpman, Melitz, and Rubinstein (2008)

argue, this ignores considerable information, potentially biasing the results. Estimation of our gravity

equation using PPML suggests the following equation:

lnXijt =exp(β0 + β1 lnGDPit + β2 lnGDPjt + β3 lnPCGDPit + β4 lnPCGDPjt + β5 lnDISTij

+ β6ADJij + β7LANGij + β8COLONYij + β9V Ajt + β10PSjt + β11GEjt + β12RQjt

+ β13RLjt + β14CCjt) + ϵijt

(2)

18

where exp denotes the exponentiated value of the term in parentheses. In our results, we estimate

equation (2) first for trade flows; we note that in our sample all trade flows have positive values.

These estimates are presented in Table 2, column (3) However, about half of our bilateral FDI flows

are zeros. Hence, to show the independent biases introduced from adjusting for heteroskedasticity

and from ignoring zeros, we first estimate equation (2) for only positively-valued FDI flows and report

these in column (4). We then estimate equation (2) using all FDI observations (positive and zero

values) and report these in column (5). Moreover, we are able to estimate equation (2) also using all

ratios of FDI to trade from i to j – even zeros – using PPML.

Table 2 reports the results. Column (3) presents the results for trade flows, which uses the same

number of observations as in Table 1, column (3). First, regarding GDPs and per capita GDPs, the

GDP elasticities fall significantly in size; this is a common outcome using PPML estimation relative to

OLS. At 0.8 for each GDP elasticity, this is in line with earlier estimates using PPML, cf., Santos Silva

and Tenreyro (2006). Also, exporter’s per capita GDP coefficient estimate, which was statistically

significant using OLS, becomes economically and statistically insignificant. Second, regarding the

time-invariant bilateral barrier variables all the coefficient estimates decrease in absolute terms, with

the common colonial ties’ coefficient estimate becoming insignificant. Third, regarding the governance

indexes, notably Voice and Accountability retains its significant negative effect on trade. Government

Effectiveness still has a positive effect on trade, but is only significant at a 10 percent level. Regulatory

Quality now has a positive and statistically significant effect on trade. We will find throughout our

empirical analysis that the key WGI indicators that influence trade flows materially are Voice and

Accountability, Government Effectiveness, and Regulatory Quality.

Columns (4) and (5) report the results for positive FDI flows only and for all FDI flows (positive

and zeros), respectively. The notable finding is that, while there is a material difference using PQML

versus OLS, there is no material difference between using PQML with just positive FDI flows versus

using PQML with all FDI flows. For brevity, we discuss only the results in column (5) using all FDI

flows. Compared with OLS, PPML yields lower coefficient estimates for home country GDP and per

capita GDP. Regarding bilateral barrier variables, distance’s, language’s and common colonial ties’

coefficient estimates are smaller in absolute terms, but adjacency’s coefficient estimate reverses sign

from positive to negative (though a negative adjacency effect for FDI is consistent with a prominent

role for horizontal FDI). Voice and Accountability ’s coefficient estimate remains insignificant, as for

OLS. Political Stability ’s coefficient estimate becomes insignificant. But Regulatory Quality ’s and Rule

of Law ’s coefficient estimates remain positive and negative, respectively, and statistically significant.

Finally, column (6) reports the coefficient estimates for FDI relative to trade. Only home and host

19

countries’ per capita GDPs, language, and common colonial ties’ coefficient estimates are positive and

statistically significant.

5.2 Fixed Effects Estimates

One of the seminal contributions of papers such as Anderson (1979), Bergstrand (1985), Eaton and

Kortum (2002), and Anderson and van Wincoop (2003) was to note the importance of accounting

for prices of other (third-party) countries in estimating bilateral gravity equations of trade flows; the

same issue is important for bilateral FDI gravity equations.28 Anderson and van Wincoop (2003), for

instance, showed elegantly how to account for “multilateral resistance” in a trade context. Since then,

many studies have accounted for the role of exporter and importer multilateral resistance (MR) terms

using exporter and importer fixed effects.

Naturally, our trade and FDI gravity equation specifications in Tables 1 and 2 are subject to the

same critique. Given our time-series of cross-sections, one method to account for the unobservable

MR terms is to introduce exporter-time and importer-time fixed effects. However, our governance

variables of interest would be suppressed because they are observed across importer/host country and

time.

In this section, we include exporter and importer fixed effects (and year dummies). However, this

is only a partial solution because the MR terms vary over time. By not including exporter-time and

importer-time fixed effects, time variation in GDPs, per capita GDPs, and the six WGI indexes can be

exploited, though the coefficient estimates for these variables may be biased by not accounting for time

variation in MR terms. Moreover, with only 5 years of annual cross-sections and most of the variation

in our sample being cross-sectional, the exporter and importer fixed effects will accordingly suppress

the influence of GDPs, per capita GDPs, and the importer/host countries’ WGIs. Consequently, we

expect few of the country-specific variables to be statistically significant given the absence of much

time variation. This motivates later in section 5.3 the use of an alternative strategy to account for

multilateral resistance terms.

Nevertheless, Table 3 presents the results using PPML and exporter and importer fixed effects;

there are several notable findings. First, regarding time-invariant bilateral variables, there are minor

changes in the coefficient estimates associated with these four variables since we are not including time-

invariant country-pair fixed effects. Although distance’s coefficient estimate increases in the trade flow

equation (column (3)) using exporter and importer fixed effects, there is no material change in the FDI

equations’ distance coefficient estimates using such fixed effects. The distinctive change for adjacency

28See Bergstrand, Gainer, and Mark (2012).

20

is that its coefficient estimate becomes insignificant in the FDI specifications. Also, language’s and

colony’s coefficient estimates using the fixed effects are now smaller in absolute terms in the trade and

FDI specifications and are often statistically insignificant. Second, regarding GDP and per capita GDP,

the notable – and not surprising – result is that GDP and per capita GDP coefficient estimates become

highly unstable, and reveal considerable multicollinearity because time variation is limited to five years.

Consequently, several of these previously low standard error estimates become high standard error

estimates. Country i’s GDP coefficient estimate is negative and large in absolute terms, but i’s per

capita GDP coefficient estimate is large and positive, confirming the high multicollinearity. Country

j’s GDP coefficient estimate is negative and j’s per capita GDP coefficient estimate is positive, also

confirming multicollinearity. Third, we surprisingly find that some of the WGI variables are significant,

even with only five years of cross-sections. Notably, for the trade gravity equation, the coefficient

estimate for Government Effectiveness is positive and statistically significant; GE was significant in

the OLS trade specification in Table 1 and even marginally significant (at 10 percent) in the PPML

specification in Table 2. For both FDI specifications in Table 3, four WGI variables are significant.

Political Stability, Regulatory Quality, and Control of Corruption are positively signed with statistical

significance; Government Effectiveness is negatively signed and statistically significant. Yet in both

FDI specifications in columns (4) and (5), none of the GDP’s or per capita GDP’s coefficient estimates

are statistically significantly different from zero given the limited time variation driving the results,

raising concern about any inferences to be drawn using exporter and importer fixed effects and only

five years of cross-sections.

5.3 Approximating Multilateral Resistance Terms

As just noted above, one of the limitations of (what has become) the standard procedure in gravity

equations to account for multilateral resistance terms with exporter and importer fixed effects is that it

precludes estimation of coefficients for non-MR terms that are exporter and importer specific. Recently,

Baier and Bergstrand (2009) showed theoretically and demonstrated both empirically and using Monte

Carlo simulations that such terms can be well approximated using the “exogenous” components of the

Anderson-van Wincoop “endogenous” non-linear MR terms. Using a first-order log-linear Taylor series

expansion, the authors showed that there are simple log-linear terms that can be incorporated into

“good old” OLS gravity equations to account for third-country prices without using non-linear least

squares estimation of the underlying structural gravity model or using exporter and importer fixed

effects. Using a comprehensive Monte Carlo analysis comparing alternative methods for estimating

gravity equations, Bergstrand, Egger and Larch (2012) show – in the absence of any measurement

21

and specification errors – that the Baier-Bergstrand approximation method yields virtually identical

coefficient and comparative static estimates to the true values.

Following Baier and Bergstrand (2009), we specify equation:

lnXijt =exp(β0 + β1 lnGDPit + β2 lnGDPjt + β3 lnPCGDPit + β4 lnPCGDPjt + β5 lnDISTij

+ β6ADJij + β7LANGij + β8COLONYij + β9V Ajt + β10PSjt + β11GEjt + β12RQjt

+ β13RLjt + β14CCjt + β15 lnMRDISTijt + β16MRADJijt + β17 lnMRLANGijt

+ β18MRCOLONYijt) + ϵijt

(3)

where MRDISTijt is the sum of a simple (or time-varying GDP-weighted) average of i’s log bilateral

distance to all other countries and a simple (or time-varying GDP-weighted) average of j’s log bilat-

eral distance to all other countries (minus a third “ROW” term), cf., Baier and Bergstrand (2009).

MRADJijt, MRLANGijt, and MRCOLONYijt are defined analogously for the adjacency, language

and colony bilateral dummy variables, respectively.

Table 4 presents the results using PPML including the MR terms described above. Column (3)

presents the results for the trade flow gravity equation using the time-varying GDP-weighted MR

terms. First, notably the inclusion of the time-varying MR terms causes the GDP coefficient esti-

mates to be more similar to traditional gravity-equation coefficient estimates, with GDP elasticities

of approximately 0.75. Per capita GDPs coefficient estimates are insignificant in the trade flow equa-

tion. Second, the four bilateral barrier variables’ coefficient estimates are similar to previous estimates

without fixed effects, with the colony variable’s coefficient estimate significant but negatively signed.

Third, for the WGI variables, Voice and Accountability and Political Stability in the importing country

have significant negative effects on trade flows. This mirrors estimates in earlier trade flows specifica-

tions, with Political Stability possibly causing less trade because it fosters more FDI, cf., Bergstrand

and Egger (2012). Government Effectiveness and Regulatory Quality have significant positive effects

on trade flows, consistent with earlier findings. Thus, the only other WGI variables – Rule of Law and

Control of Corruption have economically and statistically insignificant effects on trade flows. Finally,

we examine the MR terms in the trade equation. Based upon Baier and Bergstrand (2009), MRDIST

has the expected positive coefficient estimate and is statistically significant; the interpretation is that

– for given bilateral distance between a pair of countries – a larger “remoteness” of countries i and j

from other countries lowers the relative cost of these two countries trading, increasing their bilateral

trade flow. MRADJ has an unexpected positive coefficient sign, but is statistically insignificant.

MRLANG has the expected negative coefficient sign, but is statistically insignificant. MRCOLONY

has an unexpected positive coefficient sign, but the coefficient sign for COLONY was also unexpect-

22

edly negative; hence, the positive coefficient estimate for MRCOLONY is consistent theoretically

with the negative one for COLONY .

Columns (4) and (5) present the results for positive FDI flows and all FDI flows (positive and

zeros), respectively. First, GDP coefficient estimates have values much closer to traditional FDI gravity

equation specifications at approximately 0.80-0.84. Importantly, we find that the coefficient estimate

for home country i’s per capita GDP is positive and statistically significant, as found in Tables 1 and 2.

Human- and physical-capital abundant home countries tend to headquarter multinational firms and be

suppliers of FDI to other countries; this result is consistent with Blonigen and Piger (2011). However,

also important is that host country j’s per capita GDP is unimportant in explaining FDI flows; this is

also consistent with Tables 1 and 2. This is important because it suggests a possible variable to satisfy

the exclusion restriction for two-stage least squares estimates later using the technique described in

Helpman, Melitz, and Rubinstein (2008) to adjust gravity-equation estimates to account for (country)

selection bias alongside with accounting for firm heterogeneity. If host country per capita GDP explains

selection into FDI, but not the level of (positive) FDI, then host-country per capita GDP can be used

to “identify” the second-stage estimates using the Helpman, Melitz, Rubinstein (2008) procedure for

estimating extensive and intensive margin effects allowing for firm heterogeneity.

Second, the bilateral barrier variables have similar effects to earlier estimates in Tables 1 and 2.

Third, of all the WGI variables, Regulatory Quality has a positive and statistically significant effect

on FDI and Rule of Law has a significant negative effect in column (4) but not column (5). We will

find in the next section that results change when we allow for extensive and intensive margin effects,

as in Helpman, Melitz, and Rubinstein (2008). Finally, we note that none of the MR terms’ coefficient

estimates in columns (4) and (5) have statistically significant effects in the FDI specifications.

6 Selection into FDI and Accounting for Firm Heterogeneity

Two recent observations have inspired new methodologies associated with estimating gravity equations.

Helpman, Melitz, and Rubinstein (2008) noted that – for all the possible bilateral pairings of countries

in the world – many aggregate bilateral trade flows are zero. The same holds for aggregate bilateral

FDI flows; in our sample about half of bilateral FDI flows are zeros. Also, an entirely new literature has

surfaced in the last decade providing evidence for the heterogeneity of firms engaged internationally

and the likely importance of fixed export and FDI costs in limiting the selection of such heterogeneous

firms into international activity (either trade or FDI). While an explosion of research into the role of

firm heterogeneity and fixed export costs in explaining patterns of international trade based upon the

Melitz (2003) model has occurred, research into the implications of country selection into FDI outflows

23

and the role of firm heterogeneity for FDI levels is actually quite scant), cf., Bernard, Jensen, Redding,

and Schott (2011). In there recent survey, Bernard, Jensen, Redding, and Schott (2011) point to a

handful of papers that have explored the implications of firm heterogeneity and fixed export and FDI

costs for aggregate FDI patterns. Helpman, Melitz, and Yeaple (2004) is one of the first papers to

explore the implications of firm heterogeneity for aggregate FDI flows relative to aggregate trade flows.

Using micro data from the U.S. Bureau of Economic Analysis, Yeaple (2009) provides further evidence

supporting the Helpman, Melitz, and Yeaple (2004) theory of heterogeneous firm and FDI.

The present paper is the first to examine the effect of governance indicators for explaining the

extensive margin of FDI flows. Moreover, drawing upon the seminal methodology in Helpman, Melitz,

and Rubinstein (2008) for aggregate bilateral trade flows, we also examine the influence of governance

indicators on the intensive margin of FDI using a two-stage estimation procedure, that corrects both

for (country) sample-selection bias as well as for firm heterogeneity. Notably, we find that Voice and

Accountability – our primary measure of the pluralism component of democracy – despite having a

negative effect on the level of FDI into a country – has a significant positive effect on selection of

countries into FDI. Moreover, this result is robust to accounting for firm heterogeneity.29

As the Helpman, Melitz, and Rubinstein (2008) methodology has now become commonplace in

bilateral aggregate trade gravity equation studies, it is straightforward to apply it to bilateral aggregate

FDI flows. The framework is a two-stage estimation procedure. The first stage entails estimating a

probit equation for the determinants of the likelihood of a pair of countries having an FDI flow from

i to j. The factors that tend to determine the likelihood of a positive FDI (or trade) flow – that is,

the (country) extensive margin – tend to be the same factors that affect the level of FDI (or trade)

once it is positive – that is, the intensive margin. However, for econometric reasons, identification of

the second-stage gravity equation coefficient estimates requires at least one variable that significantly

explains the likelihood of FDI that does not affect the level of FDI, i.e., the intensive margin; this

is the “exclusion restriction.” Our results earlier, however, suggest a likely candidate variable is the

host country j ’s per capita GDP. Examination of FDI gravity equation results in Tables 1-4 reveals

consistently that – while home country i’s per capita GDP is a systematically important determinant

of FDI levels – host country j’s per capita GDP coefficient estimate is never statistically significant,

and in most cases (Tables 1, 2, and 4) not economically significant either. This differential influence

on FDI flows between home and host country per capita GDPs is consistent with findings in Blonigen

and Piger (2011) as well. As we will see, however, host country j’s per capita GDP is an economically

29Egger, Larch, Staub, and Winkelmann (2011) have recently emphasized the importance also of country selectionbias and (to a lesser extent) firm heterogeneity for influencing bilateral aggregate trade flows, as well as the importanceof accounting for the endogeneity of regional trade agreement dummies.

24

and statistically significant determinant of the likelihood of a positive FDI flow from i to j.30

We summarize the econometric procedure briefly, referring the reader to Helpman, Melitz, and

Rubinstein (2008) for details. The first-stage entails estimating a probit equation for FDI flows

using the same RHS variables as used in earlier specifications, including country j’s per capita GDP.

With these results, the predicted probabilities (denoted ρijt) are calculated. The ρijt can be used

along with the cumulative normal distribution function, Φ(), to form estimates zijt where zijt =

Φ−1(ρijt). Estimates of zijt are necessary to construct two variables for use in the second-stage gravity

equation, ηijt (which will control for selection bias) and wijt (which will control for bias associated

with “firm heterogeneity”). Using the zijt, we can estimate the Inverse Mill’s Ratio, ηijt, where

ηijt = φ(zijt)/Φ(zijt) and φ() is the normal density function. The ηijt are then included in the second-

stage gravity equation to control for selection bias. Following Helpman, Melitz, and Rubinstein (2008),

or HMR, the zijt are also used to construct a variable wijt, where wijt = ln exp[δ(zijt + ηijt)]− 1, which

in the context of the HMR model can be included in the second-stage estimation to control for bias

associated with the influence of firm heterogeneity.

Table 5 presents the results of estimating the first-stage probits – under various specifications

– and second-stage HMR procedure using column (5) first-stage results. Columns (3), (4) and (5)

provide alternative probit estimates; because probits are non-linear we report the “marginal” effects

(conditioned on the means), so these estimates have economic significance (unlike the actual probit

coefficients). Column (3) presents the probit estimates using the traditional RHS variables, such as

used in Table 1 earlier. Column (4) presents marginal logit, rather than probit, estimates including

exporter and importer fixed effects. Column (5) presents the probit estimates including instead the

MR (approximation) terms discussed earlier.31

We examine Column (3) by the three categories of variables, as done earlier. First, we find that all

four GDP and per capita GDP variables are statistically significant. Home and host country GDPs

have positive and significant marginal effects, as well as home country per capita GDP. Importantly

also is that host country j’s per capita GDP coefficient estimate is statistically significant; this will be

important for identification in the second-stage regression. Second, all the bilateral barrier variables

have expected signs for their marginal effects, and those for distance, adjacency, and common colony

are statistically significant. Third, four of the six WGI variables’ coefficient estimates are statistically

significant. Notably, Voice and Accountability has a positive and statistically significant marginal effect

30Helpman, Melitz, and Rubinstein (2008) used religion and regulatory costs variables to satisfy the exclusion restric-tion.

31For the specification including exporter and importer fixed effects, coefficient estimates in probits with such fixedeffects can lead to biased results, cf., Wooldridge (2010), leading researchers instead to use the alternative Chamberlainapproach. To avoid these issues, we use logit to determine marginal effects in the presence of exporter and importerfixed effects.

25

on the likelihood of FDI (Prob(FDI)). Political Stability and Regulatory Quality also have positive

and significant marginal effects on the probability of FDI. Finally, Rule of Law has a statistically

significant negative effect on the probability of FDI; however, once we include either fixed effects or

MR terms, this marginal effect becomes insignificant.

Column (4) reports the logit marginal effects including exporter and importer fixed effects. Because

the exporter and importer fixed effects subsume the cross-sectional variation in GDPs, per capita

GDPs, and the WGIs, we are not optimistic about the results. Not surprisingly, as was the case in

Table 3 for FDI values, the model performs poorly, having to rely only upon five years for time-series

variation. Nevertheless, we report and summarize the results. First, while home country GDP’s

marginal effect estimate is significant, host country GDP’s marginal effect estimate is not; the same

holds for home and host countries’ per capita GDPs. The marginal effects for distance, adjacency,

language, and colony are correctly signed and significant. However, none of the WGI variables’

marginal effects are statistically significant in column (4). The absence of statistical significance for

country j’s per capita GDP implies that second-stage estimates of FDI levels using exporter and

importer fixed effects are likely infeasible, due to the absence of a significant identifying variable for

the second stage.

Column (5) reports the probit results using the (exogenous) time-varying MR (approximation)

terms discussed earlier in section 5.3. The results using the MR terms are similar to those in column

(3), which exclude them. First, home and host countries’ GDPs have significant positive marginal

effects on Prob(FDI), as does home country per capita GDP. Also, host country per capita GDP has

a significant effect on Prob(FDI), which is important for identification in the second stage. Distance,

adjacency, and colony have significant effects on Prob(FDI), but of the MR terms only MRADJ

has a significant effect. As in column (3), Voice and Accountability has a significant positive effect on

Prob(FDI), along with Political Stability and Regulatory Quality. However, whereas in column (3)

Rule of Law had a significant negative impact on Prob(FDI), in column (5) Rule of Law no longer

has a significant impact on Prob(FDI).

Finally, column (6) reports the results of estimating the second stage regression using the HMR

methodology and omitting host country j’s per capita GDP, consistent with earlier evidence that this

variable has no material impact on FDI levels. First, as before home and host GDPs have significant

positive effects on FDI. Importantly, unlike corresponding estimates in Tables 2 and 4, the home

country GDP elasticity is larger than the host country GDP elasticity – as theory in Bergstrand and

Egger (2007) suggests. Home per capita GDP elasticity is positive and significant, as earlier. Second,

regarding the time-invariant bilateral variables distance has the expected negative relationship with

26

FDI, while adjacency, language and colony have the expected positive relationships. While adjacency’s

coefficient estimate is similar to earlier ones, the standard error is large enough to preclude significance;

however, both the language and colony dummies’ coefficient estimates are statistically significant.

Third, of the WGI variables, as earlier Voice and Accountability is negatively related to FDI levels and

remains statistically insignificant at the 5 percent level, but is significantly negative at the 10 percent

level. Political Stability and Regulatory Quality are not just positively related significantly to the

probability of FDI, but also to the level of FDI. Rule of Law still has an unexpected positive relationship

with FDI, and is significant at 5 percent. Fourth, MRDIST now has a positive and significant

relationship with the level of FDI, in contrast with earlier results; this is consistent theoretically with

the negative effect for bilateral distance. MRADJ and MRCOLONY both have negative effects with

FDI, as one might expect; however, neither variable’s coefficient estimate is significant. MRLANG

has a positive and significant effect on FDI, which is inconsistent with the positive effect for the

bilateral language dummy variable. Finally, similar to Egger, Larch, Staub, and Winkelmann (2011)

for trade flows, the Heckman selection variable, η, has a positive and statistically significant effect and

the index for firm heterogeneity, δ, has no significant effect. Consequently, adjustment for selection

bias is important indicating the relevance of the two-stage estimation procedure for levels. However,

potential bias from firm heterogeneity likely has little empirical relevance.

7 Conclusions

Unlike the large literature on “democracy and trade,” there is a much smaller literature on the effect

of the level of democracy in a nation on the level of foreign direct investment (FDI). Moreover, of these

few studies, only one has used a gravity-equation methodology to explore the impact of democracy on

bilateral FDI flows, many studies have not explored the potentially conflicting effects of the numerous

dimensions of democracy on FDI, and no study has explored how different dimensions of democracy

influence the selection of countries into bilateral FDI outflows to host countries.

This study has focused on three potential contributions. First, we examined the effects of the

(six) World Bank’s Worldwide Governance Indicators (WGIs) – which allow separating the effects

of pluralism from those of five other good governance measures – on bilateral trade, FDI, and FDI-

relative-to-trade flows using state-of-the-art gravity specifications motivated by the general equilibrium

Knowledge-and-Physical-Capital model in Bergstrand and Egger (2007, 2010, 2012). Second, we found

strong evidence that – after accounting for host governments’ effectiveness in various roles of good

governance – a higher level of pluralism as measured by the WGIs’ Voice and Accountability index

reduces trade levels, likely by increasing the “voice” of more protectionist less-skilled workers, but

27

not FDI levels. Moreover, we find qualitatively different effects of several WGIs – such as political

stability and government effectiveness – on trade versus FDI. Third, we account for firm heterogeneity

alongside a large number of zeros in bilateral FDI flows using recent advances in gravity modeling.

We distinguish between the intensive and extensive margins and show that pluralism (as measured

by Voice and Accountability) affects FDI inflows negatively at the (country) intensive margin but

positively at the extensive margin.

Future research needs to explore in more detail the effects of different dimensions of democracy on

selection into versus levels of trade and FDI. In aiming to include measures of democratic institutions

computed systematically – the well known World Bank WGIs – we have had to forego a longer time

dimension in our pooled cross-section time-series analysis. Further work needs to explore in a more

comprehensive panel the effects of such institutional dimensions on trade and FDI, and subsequently

their impacts on economic growth.

References

Adam, Antonis, and Fragkiskos Filippaios (2007), Foreign direct investment and civil liberties: A new

perspective, European Journal of Political Economy 23 (4), 1038-1052.

Alcala, Francisco, and Antonio Ciccone (2004), Trade and productivity, Quarterly Journal of Eco-

nomics 119 (2), 612-645, May.

Alesina, Alberto, and David Dollar (2000), Who gives foreign aid to whom and why?, Journal of

Economic Growth 5 (10, 33-63.

Anderson, James E. (1979), A theoretical foundation for the gravity equation, American Economic

Review 69, 106-16.

Anderson, James E. (2011), The gravity model, Annual Review of Economics, forthcoming.

Anderson, James E., and Eric van Wincoop (2003), Gravity with gravitas: a solution to the border

puzzle, American Economic Review 93, 170-92.

Anderson, James E., and Eric van Wincoop (2004), Trade costs, Journal of Economic Literature 42,

691-751.

Arkolakis, Costas, Arnaud Costinot, and Andres Rodriguez-Clare (2011), New trade models, same old

gains? American Economic Review 102, 94-130.

28

Arndt, Christiane, and Charles Oman (2006), Uses and abuses of governance indicators, Organization

for Economic Cooperation and Development (OECD) Development Center Study.

Asiedu, Elizabeth, and Donald Lien (2011), Democracy, foreign direct investment, and natural re-

sources, Journal of International Economics 84 (1), 99-111.

Badinger, Harald (2008), Trade policy and productivity, European Economic Review 52 (5), 867-891,

July.

Baier, Scott L., and Jeffrey H. Bergstrand, 2007. Do free trade agreements actually increase members’

international trade? Journal of International Economics 71, 72-95.

Baier, Scott L., and Jeffrey H. Bergstrand (2009), Bonus vetus OLS: A simple method for approximat-

ing international trade-cost effects using the gravity equation, Journal of International Economics

77, 77-85.

Benassy-Quere, Agnes, Maylis Coupet, and Thierry Mayer (2007), Institutional determinants of foreign

direct investment, The World Economy 30 (5), 764-782.

Bergstrand, Jeffrey H. (1985), The gravity equation in international trade: Some microeconomic

foundations and empirical evidence, Review of Economics and Statistics 67 (3), 474-481.

Bergstrand, Jeffrey H., and Peter Egger (2007), A knowledge-and-physical-capital model of interna-

tional trade flows, foreign direct investment, and multinational enterprises, Journal of International

Economics, 73, 278-308.

Bergstrand, Jeffrey H., and Peter Egger (2010), A general equilibrium theory for estimating grav-

ity equations of bilateral FDI, final goods trade, and intermediate trade flows. In Peter A.G.

van Bergeijk and Steven Brakman (eds.), The Gravity Model in International Trade. Cambridge

University Press, Cambridge, UK.

Bergstrand, Jeffrey H., and Peter Egger (2011a), Shouldn’t physical capital also matter for multina-

tional enterprise activity?, unpublished manuscript, University of Notre Dame.

Bergstrand, Jeffrey H., and Peter Egger (2011b), Gravity equations and economic frictions in the

world economy. In D. Bernhofen, R. Falvey, and U. Kreickemeier (eds.), Palgrave Handbook of

International Trade, London, UK: Macmillan Palgrave.

Bergstrand, Jeffrey H., and Peter Egger (2012), What determines BITs?, unpublished manuscript,

University of Notre Dame, January.

29

Bergstrand, Jeffrey H., Peter Egger, and Mario Larch (2012), Gravity redux : Estimation of gravity-

equation coefficients, elasticities of substitution, and general equilibrium comparative statics under

asymmetric bilateral trade costs, unpublished manuscript, University of Notre Dame, September.

Bergstrand, Jeffrey H., Mitch Gainer, and Nelson Mark (2012), Structural gravity for foreign direct

investment, unpublished manuscript, University of Notre Dame.

Bernard, Andrew B, J. Bradford Jensen, Stephen J. Redding, and Peter K. Schott (2011), The empirics

of firm heterogeneity and international trade, National Bureau of Economic Research Working

Paper No. 17627, November.

Blonigen, Bruce A. (2005), A review of the empirical literature on FDI determinants, Atlantic Eco-

nomic Journal 33 (4), 383-403.

Blonigen, Bruce A., Ronald B. Davies, and Keith Head (2003), Estimating the knowledge-capital

model of the multinational enterprise: comment, American Economic Review 93 (3), 980-994.

Blonigen, Bruce A., and Jeremy Piger (2011), Determinants of foreign direct investment, National

Bureau of Economic Research Working Paper No. 16704, January.

Busse, Matthias (2004), Transnational corporation and repression of political rights and civil liberties:

An empirical analysis, Kyklos 57 (1), 45-66.

Busse, Matthias, and Carsten Hefeker (2007), Political risk, institution and foreign direct investment,

European Journal of Political Economy 23 (2), 397-415.

Buthe, Tim, and Helen V. Milner (2008), The politics of foreign direct investment into developing

countries Increasing FDI through international trade agreements?, American Journal of Political

Science 52 (4), 741-762.

Carr, David, James R. Markusen, and Keith E. Maskus (2001), Estimating the knowledge-capital

model of the multinational enterprise, American Economic Review, 91 (3), June, 693-708.

Eaton, Jonathan, and Samuel Kortum (2002), Technology, geography, and trade, Econometrica 70

(5), 1741-1779.

Egger, Peter, Mario Larch, Kevin E. Staub, and Rainer Windelmann (2011), The trade effects of en-

dogenous preferential trade agreements, American Economic Journal: Economic Policy 3, August,

113-143.

30

Eicher, Theo, Lindy Helfman, and Alex Lenkoski (2011), Robust FDI determinants: Bayesian model

averaging in the presence of selection bias, unpublished manuscript, November.

Globerman, Steven, and Daniel Shapiro (2002), Global foreign direct investment flows: The role of

governance infrastructure, World Development 30 (11), 1899-1919.

Globerman, Steven, and Daniel Shapiro (2003), Governance infrastructure and U.S. foreign direct

investment, Journal of International Business 34 (1), January, 19-39.

Harms, Philipp, and Heinrich W. Ursprung (2002), Do civil and political repression really boost foreign

direct investment?, Economic Inquiry 40 (4), October, 651-663.

Helpman, Elhanan (1984), A simple theory of trade with multinational corporations, Journal of Po-

litical Economy 92, 451-471

Helpman, Elhanan, and Paul Krugman (1985), Market Structure and Foreign Trade, MIT Press,

Cambridge, Massachusetts.

Helpman, Elhanan, Marc J. Melitz, and Yona Rubinstein (2008), Estimating trade flows: trading

partners and trading volumes, Quarterly Journal of Economics 123 (2), 441-487.

Helpman, Elhanan, Marc J. Melitz, and Stephen R. Yeaple (2004), Export versus FDI with heteroge-

neous firms, American Economic Review 94 (1), March, 300-316.

Jakobsen, Jo (2006), Does democracy moderate the obsolescing bargain mechanism? An empirical

analysis, 1983-2001, Transnational Corporations 15 (3), 67-106.

Jakobsen, and Indra de Soysa (2006), Do foreign investors punish democracy? Theory and empirics,

1984-2001, Kyklos 59 (3), 383-410.

Jensen, Nathan M. (2003), Democratic governance and multinational corporations: Political regimes

and inflows of foreign direct investment, International Organization 57, 587-616.

Kaufmann, Daniel, Aart Kraay, and Massimo Mastruzzi (2006), Growth and governance: A reply,

World Bank, July.

Kaufmann, Daniel, Aart Kraay, and Massimo Mastruzzi (2007a), Governance matters VI: Aggregate

and individual governance indicators, 1996-2006, World Bank Policy Research Working Paper No.

4280, July.

Kaufmann, Daniel, Aart Kraay, and Massimo Mastruzzi (2007b), the worldwide governance indicators

project: Answering the critics, World Bank, February.

31

Kaufmann, Daniel, Aart Kraay, and Massimo Mastruzzi (2007c), Growth and governance: A rejoinder,

Journal of Politics 69 (2), May, 570-572.

Kurtz, Marcus J., and Andrew Schrank (2007), Journal of Politics 69 (2), May, 538-554.

Li, Quan, and Adam Resnick (2003), Reversal of fortunes: Democratic institutions and foreign direct

investment inflows to developing countries, International Organization 57 (1), 175-211.

Li, Quan, and Rafael Reuveny (2003), Economic globalization and democracy: An empirical analysis,

British Journal of Political Science 33, 29-54.

Linnemann, Hans (1966), An Econometric Study of International Trade Flows, North-Holland Pub-

lishing, Amsterdam.

Markusen, James R. (1984), Multinationals, multi-plant economies, and the gains from trade, Journal

of International Economics 16 205-226.

Markusen, James R. (2002), Multinational Firms and the Theory of International Trade, MIT Press,

Cambridge, Massachusetts.

Markusen, James R., and Keith E. Maskus (2001), Multinational firms: Reconciling theory and evi-

dence. In Magnus Blomstron and Linda Goldberg (eds.), Topics in Empirical International Eco-

nomics: A Festschrift in Honor of Robert E. Lipsey, University of Chicago Press, Chicago, IL.

Markusen, James R., and Keith E. Maskus (2002), Discriminating among alternative theories of the

multinational enterprise, Review of International Economics, 10, no. 4, 694-707.

Markusen, James R., and Anthony J. Venables (1998), Multinational firms and the new trade theory,

Journal of International Economics 46, 183-203.

Markusen, James R., and Anthony J. Venables (2000), The theory of endowment, intra-industry, and

multinational trade, Journal of International Economics 52, 209-235.

Mayda, A.M., and Dani Rodrik (2005), Why are some individuals (and countries) more protectionist

than others?, European Economic Review 49, 1393-1430.

Melitz, Marc J. (2003), The impact of trade on intra-industry reallocations and aggregate industry

productivity, Econometrica 71 (6), November, 1695-1725.

Milner, Helen V., and Bumba Mukherjee (2009), Democratization and economic globalization, Annual

Review of Political Science 12, 163-181.

32

O’Neal, John R. (1994), The affinity of foreign investors for authoritarian regimes, Political Research

Quarterly 47, 565-588.

Ramondo, Natalia, and Andres Rodriguez-Clare (2009), Trade, multinational production, and the

gains from openness, National Bureau of Economic Research Working Paper No. 15604, December.

Rodrik, Dani (1996), Labor standards in international trade: Do they matter and what do we about

them?, In Robert Lawrence, Dani Rodrik, and John Whalley (eds.), Emerging Agenda for global

Trade: High Stakes for Developing Countries, Johns Hopkins University Press, Baltimore, Mary-

land.

Santos Silva, J.M.C., and Silvana Tenreyro (2006), The log of gravity, Review of Economics and

Statistics 88, 641-658.

Scheve, K.F., and Matthew J. Slaughter (2001), What determines individual trade policy preferences?,

Journal of International Economics 54, 267-292.

Wooldridge, Jeffrey M. (2010), Econometric Analysis of Cross Section and Panel Data, 2nd Edition,

MIT Press, Cambridge, Massachusetts.

Yeaple, Stephen R. (2009), Firm heterogeneity and the structure of U.S. multinational activity, Journal

of International Economics 78 (2), July, 206-215.

33

Table 1: OLS

(1) (2) (3) (4) (5)

LHS

Variables

Expected

Coefficient

Signs for

ijt ijt ijt ijtCols. 3 & 4 ln TRADE ln FDI ln (FDI /TRADE )

itln GDP + 0.93*

(49.98)

0.91*

(23.72)

0.14*

(4.20)

jtln GDP + 0.91*

(24.40)

0.74*

(10.46)

0.04

(0.63)

itln PCGDP ? 0.38*

(9.63)

1.40*

(14.10)

1.79*

(18.43)

jtln PCGDP ? -0.01

(-0.08)

0.01

(0.12)

-0.05

(-0.37)

ijln DIST - -1.07*

(-22.73)

-0.52*

(-6.26)

0.22*

(2.70)

ijADJ + / ? 0.20

(1.16)

0.88*

(3.07)

0.61*

(2.14)

ijLANG + 0.67*

(5.08)

1.08*

(5.77)

0.52*

(3.22)

ijCOLONY + 0.60*

(4.69)

0.60*

(3.40)

0.40*

(2.24)

jtVA ? -0.012*

(-2.68)

-0.007

(-1.05)

0.008

(1.52)

jtPS + -0.001

(-0.32)

0.010*

(2.05)

0.008

(1.85)

jtGE + 0.012*

(2.85)

0.007

(0.68)

-0.007

(-0.66)

jtRQ + 0.008

(1.73)

0.023*

(2.38)

0.018

(1.77)

jtRL + -0.003

(-0.46)

-0.023*

(-2.16)

-0.017

(-1.85)

jtCC + -0.000

(-0.00)

-0.000

(-0.03)

-0.003

(-0.41)

Constant -12.97*

(-17.20)

-26.38*

(-19.97)

-23.35*

(-18.10)

N 12,229 4,840 4,840

R 0.83 0.66 0.412

RMSE 1.30 1.67 1.70

F-stat. 419.94 160.10 38.08

*Denotes statistically significant in two-tail t-test at the 5 percent level using clustered standard errors (clustered on

destination variables).

Table 2: PPML

(1) (2) (3) (4) (5) (6)

LHS

Variables

Expected

Coefficient

Signs

ijt ijtfor Cols. 3, 4, 5 TRADE >0 FDI >0 ijt ijt ijtFDI FDI / TRADE

itln GDP + 0.81*

(26.77)

0.79*

(13.16)

0.81*

(13.81)

-0.04

(-0.96)

jtln GDP + 0.82*

(16.23)

0.78*

(11.24)

0.80*

(12.19)

-0.17

(-0.95)

itln PCGDP ? -0.09

(-1.05)

0.91*

(9.41)

0.95*

(9.55)

2.07*

(4.22)

jtln PCGDP ? -0.12

(-1.46)

-0.04

(-0.18)

-0.03

(-0.16)

0.80*

(2.48)

ijln DIST - -0.69*

(-18.20)

-0.60*

(-5.62)

-0.63*

(-6.01)

0.39

(1.80)

ijADJ + / ? 0.48*

(3.90)

-0.45

(-1.85)

-0.48*

(-2.00)

-0.02

(-0.04)

ijLANG + 0.32*

(4.19)

0.84*

(6.92)

0.85*

(7.00)

0.85*

(3.11)

ijCOLONY + -0.17

(-1.01)

0.34*

(2.88)

0.34*

(2.82)

1.36*

(3.94)

jtVA ? -0.017*

(-3.61)

-0.001

(-0.17)

-0.0005

(-0.07)

0.006

(0.68)

jtPS + -0.003

(-1.63)

0.003

(0.60)

0.004

(0.74)

0.003

(0.37)

jtGE + 0.012

(1.94)

0.015

(0.78)

0.018

(0.95)

0.029

(1.46)

jtRQ + 0.030*

(4.61)

0.047*

(3.36)

0.047*

(3.49)

-0.044

(-1.91)

jtRL + -0.002

(-0.33)

-0.048*

(-2.07)

-0.047*

(-2.02)

-0.024

(-1.80)

jtCC + -0.006

(-0.77)

0.019

(1.11)

0.015

(0.87)

-0.004

(-0.32)

Constant -7.54*

(-10.35)

-20.01*

(-12.74)

-21.00*

(-14.42)

-27.06*

(-4.62)

N 12,229 4,840 9,360 9,200

*Denotes statistically significant in two-tail t-test at the 5 percent level using clustered standard errors (clustered on

destination variables). Pseudo-R is not applicable.2

Table 3: PPML with Fixed Effects

(1) (2) (3) (4) (5) (6)

LHS

Variables

Expected

Coefficient

Signs

ijt ijt ijt ijtfor Cols. 3, 4, 5 TRADE >0 FDI >0 FDI FDI/TRADE

itln GDP + -2.41*

(-3.16)

1.74

(1.12)

2.13

(1.45)

17.75*

(3.14)

jtln GDP + -0.48

(-1.05)

-1.18

(-1.20)

-1.06

(-1.09)

-3.00*

(-2.08)

itln PCGDP ? 3.09*

(4.09)

-1.18

(-0.73)

-1.52

(-0.99)

-16.51*

(-3.07)

jtln PCGDP ? 1.18

(2.73)

1.52

(1.52)

1.36

(1.36)

2.91*

(2.01)

ijln DIST - -0.90*

(-20.58)

-0.55*

(-11.11)

-0.56*

(-10.73)

0.30*

(3.84)

ijADJ + / ? 0.30*

(3.10)

-0.07

(-0.55)

-0.10

(-0.77)

-0.08

(-0.24)

ijLANG + 0.17

(1.58)

0.48*

(3.21)

0.49*

(3.30)

0.61*

(3.56)

ijCOLONY + -0.20

(-1.58)

0.27

(1.84)

0.27

(1.85)

0.66*

(3.41)

jtVA ? -0.002

(-0.99)

0.001

(0.29)

0.001

(0.28)

0.012*

(2.12)

jtPS + 0.001

(1.06)

0.004*

(4.39)

0.004*

(4.25)

0.002

(0.080)

jtGE + 0.012*

(6.24)

-0.015*

(-3.96)

-0.016*

(-3.07)

-0.026*

(-3.29)

jtRQ + 0.001

(0.63)

0.008*

(2.38)

0.008*

(2.40)

0.003

(0.56)

jtRL + 0.002

(0.47)

-0.009

(-1.11)

-0.007

(-1.03)

-0.018

(-1.72)

jtCC + -0.007

(-0.96)

0.009*

(2.20)

0.009*

(2.26)

0.000

(0.04)

Constant 9.43*

(2.33)

-7.98

(-0.97)

-13.41

(-1.75)

-19.13

(-1.04)

Source Country FE Yes Yes Yes Yes

Destination Country FE Yes Yes Yes Yes

Year FE Yes Yes Yes Yes

N 12,229 4,840 9,360 9,200

*Denotes statistically significant in two-tail t-test at the 5 percent level using clustered standard errors (clustered on

destination variables). Pseudo-R is not applicable.2

Table 4: PPML with “Approximated” MR Terms

(1) (2) (3) (4) (5) (6)

LHS

Variables

Expected

Coefficient

Signs

ijt ijtfor Cols. 3, 4, 5 TRADE >0 FDI >0

ijt ijt ijtFDI FDI /TRADE

itln GDP + 0.74*

(33.44)

0.80*

(8.42)

0.82*

(8.66)

0.03

(0.72)

jtln GDP + 0.78*

(18.18)

0.82*

(10.07)

0.84*

(10.57)

-0.06

(-0.44)

itln PCGDP ? 0.02

(0.29)

0.86*

(6.89)

0.90*

(7.01)

1.77*

(5.88)

jtln PCGDP ? -0.04

(-0.49)

-0.07

(-0.29)

-0.05

(-0.24)

0.91*

(2.90)

ijln DIST - -0.94*

(-23.73)

-0.65*

(-7.37)

-0.68*

(-8.00)

0.74

(1.78)

ijADJ + / ? 0.43*

(4.56)

-0.27

(-1.72)

-0.32*

(-2.04)

0.77

(0.76)

ijLANG + 0.23*

(3.05)

0.59*

(4.08)

0.61*

(4.23)

0.25

(1.59)

ijCOLONY + -0.23*

(-2.12)

0.25

(1.66)

0.25

(1.65)

1.23*

(5.19)

jtVA ? -0.009*

(-2.78)

0.003

(0.34)

0.004

(0.37)

0.027*

(2.62)

jtPS + -0.004*

(-2.24)

0.007

(1.03)

0.008

(1.11)

0.011

(1.20)

jtGE + 0.009

(1.40)

0.011

(0.62)

0.014

(0.82)

0.028

(1.94)

jtRQ + 0.022*

(5.45)

0.041*

(2.98)

0.043*

(3.15)

-0.051

(-1.84)

jtRL + -0.006

(-0.76)

-0.053*

(-1.98)

-0.052

(-1.94)

-0.041*

(-2.27)

jtCC + 0.001

(0.07)

0.021

(1.19)

0.016

(0.94)

-0.007

(-0.44)

ijtln MRDIST + 0.71*

(9.97)

-0.12

(-0.44)

-0.10

(-0.35)

-1.16

(-1.66)

ijtln MRADJ - / ? 1.10

(0.85)

-5.38

(-0.92)

-4.65

(-0.79)

-14.30

(-1.27)

ijtln MRLANG - -0.26

(-1.11)

1.24

(1.87)

1.24

(1.86)

1.69*

(2.57)

ijtln MRCOLONY - 3.06*

(2.96)

1.87

(0.89)

1.62

(0.78)

6.76

(1.59)

Constant -11.68*

(-12.06)

-17.85*

(-6.24)

-19.12*

(-6.95)

-19.91*

(-7.26)

N 12,229 4,840 9,360 9,200

*Denotes statistically significant in two-tail t-test at the 5 percent level using clustered standard errors (clustered on

destination variables). Pseudo-R is not applicable.2

Table 5: Probit Regressions for FDI (Marginal Effects) and Second-Stage HMR Regression

(1) (2) (3) (4) (5) (6)

LHS

Variables

Expected Coeff.

Signs for

ijtCols. 3, 4, 5, 6 Prob (FDI )

ijt Prob (FDI )

with

Fixed EffectsaijtProb (FDI )

with MR Terms

Second-Stage

HMR

FDI > 0

itln GDP + 0.11*

(12.81)

2.72*

(3.25)

0.10*

(11.16)

0.86*

(17.07)

jtln GDP + 0.20*

(14.76)

-0.18

(-0.37)

0.19*

(13.64)

0.72*

(8.03)

itln PCGDP ? 0.19*

(8.83)

-2.78*

(-3.37)

0.21*

(10.11)

1.33*

(10.94)

jtln PCGDP ? -0.07*

(-3.52)

0.10

(0.21)

-0.06*

(-2.77)

—

ijln DIST - -0.20*

(-12.22)

-0.57*

(-6.18)

-0.18*

(-6.10)

-0.82*

(-8.68)

ijADJ ? 0.41*

(8.95)

0.50*

(7.25)

0.41*

(3.92)

0.37

(1.14)

ijLANG + 0.03

(0.53)

0.38*

(5.55)

0.07

(1.11)

0.57*

(2.84)

ijCOLONY + 0.25*

(4.19)

0.40*

(5.12)

0.23*

(3.37)

0.65*

(2.77)

jtVA ? 0.002*

(2.25)

-0.002

(-0.68)

0.002*

(1.98)

-0.011

(-1.73)

jtPS + 0.004*

(3.32)

-0.001

(-0.47)

0.003*

(2.86)

0.011*

(2.47)

jtGE + -0.001

(-0.49)

-0.002

(-0.68)

-0.001

(-0.54)

0.017

(1.82)

jtRQ + 0.005*

(4.27)

0.004

(1.81)

0.005*

(4.11)

0.022*

(2.73)

jtRL + -0.003*

(-1.99)

0.003

(0.77)

-0.003

(-1.53)

-0.022

(-1.88)

jtCC + -0.001

(-0.40)

0.002

(0.91)

-0.001

(-0.25)

-0.009

(-1.02)

ijtln MRDIST + -0.02

(-0.29)

0.57*

(2.99)

ijtln MRADJ ? 1.94*

(2.87)

-3.80

(-1.40)

ijtln MRLANG - -0.03

(-0.26)

1.49*

(2.40)

ijtln MRCOLONY - -0.25

(-0.95)

-0.87

(-0.73)

ç estimate

(Selection)

1.37*

(4.92)

ä estimate

(Firm Heterog.)

0.07

(0.38)

Constant -25.09*

(-9.80)

Source Country FE No Yes No No

Destination Country FE No Yes No No

Year FE No Yes No No

N 9,360 8,488 9,360 4,840

Pseudo-R 0.47 0.71 0.47 na2

*Denotes statistically significant in two-tail z-test at the 5 percent level using clustered standard errors (clustered on

destination variables). Column 4 marginal effects are based on logit, not probit; see text for explanation.a