How trade and investment agreements affect bilateral foreign direct investment: Results from a structural gravity model

• Date: 01 December 2020
• DOI: http://doi.org/10.1111/twec.13002
• Published date: 01 December 2020
• Author: Hugo Rojas‐Romagosa,Henk L. M. Kox

World Econ. 2020;43:3203–3242. wileyonlinelibrary.com/journal/twec

|

3203

© 2020 John Wiley & Sons Ltd

Received: 17 September 2019

|

Revised: 28 March 2020

|

Accepted: 4 June 2020

DOI: 10.1111/twec.13002

ORIGINAL ARTICLE

How trade and investment agreements affect

bilateral foreign direct investment: Results from a

structural gravity model

Henk L. M.Kox1

|

HugoRojas-Romagosa2

1KVL Economic Policy Research, Den Bosch, The Netherlands

2World Trade Institute, Universität Bern, Bern, Switzerland

KEYWORDS

foreign direct investment, preferential trade agreements, gravity model

1

|

INTRODUCTION

We estimate the potential impact of preferential trade agreements (PTAs)—and other bilateral policies

that affect trade and investment—on the bilateral FDI stocks and flows between the countries signing

these agreements. Our key results are based on a structural gravity model of FDI that is applied to

bilateral FDI data from UNCTAD (2014), while controlling for the presence, heterogeneity and depth

of preferential trade agreements (PTAs), and other time-varying bilateral policies.

Starting in the 1990s, the world economy has experienced a large increase in the number of PTAs,

and also in their ‘depth’—measured by the number of their operative provisions.1 While stimulating

bilateral trade is the main focus of most PTAs, recent preferential trade agreements increasingly con-

tain provisions on bilateral investment between member states. However, the impact of PTAs on the

magnitude of foreign direct investments is not straightforward. Trade and FDI can either complement

or substitute each other, depending on the investment motivation (i.e., horizontal and vertical), the

specific industry and on the way in which the FDI provisions are shaped in the PTA. From a theoreti-

cal point of view, horizontal FDI—where firms replicate domestic activities in a foreign country—is

associated with FDI substituting for trade. Thus, in the presence of horizontal FDI, PTAs are expected

to decrease FDI flows. On the other hand, vertical FDI—where firms split activities between different

geographical regions—creates a complementary relationship between trade, PTAs and FDI (cf.

Markusen,2002). More recently, and, in part, due to the expansion and complexity of global value

chains (GVCs), other motives for FDI have been identified. For instance, export-platform FDI where

MNEs produce to export to third markets (Ekholm, Forslid, & Markusen,2007; Hanson, Mataloni, &

Slaughter,2005). Yeaple (2003) classifies mixed FDI motives as ‘complex FDI’. Baldwin and Okubo

1See for example, Dür et al. (2014), Hofmann et al. (2017), and Miroudot and Rigo (2019).

3204

|

KOX and ROJaS-ROMaGOSa

(2014) develop the concepts of ‘horizontal-ness’ and ‘vertical-ness’ to systematically classify these

more complex forms of FDI. Horizontal-ness is related to large shares of local sales, while verti-

cal-ness is associated with large shares of local sourcing of intermediates.2 The links between trade

and FDI, especially the role of complementarity (vertical-ness) and substitutability (horizontal-ness),

most probably differ by industry. Hence, the industry composition of bilateral trade and FDI is also

likely to affect how PTAs impact bilateral FDI.

There has also been an increasing gap between the refinement of theoretical FDI models and the

availability of bilateral FDI data. The motivation or purpose of the investments is mostly not revealed,

and it can only be inferred in an indirect way. Industry decomposition of bilateral FDI is only possible

for a very small number of countries.3 Even the compilation of a consistent set of total bilateral FDI

data between country-pairs and at world level is a difficult task (IMF, 2003). It is common that two

partner countries report different bilateral FDI values. Thus, a detailed reconciliation of the data and

a compilation methodology is required. In this study, we employ such a compilation, done by the

UNCTAD (2014) database, which provides bilateral data on inward and outward FDI flows and

stocks.4

Since we cannot separate the FDI data between horizontal and vertical FDI, the expected relation

between PTAs, trade and FDI remains an empirical question. Bergstrand and Egger (2007) test a

model with three countries with export and FDI, and they find that regional integration has a positive

trade effect, but a negative effect on FDI. This suggests a substitutability relation between trade and

FDI. Other studies that employ panel data with a larger set of countries find the opposite results

(Anderson, Larch, & Yotov,2019, 2020; Chen,2009; Daude, Levy Yeyati, & Stein, 2003; Osnago,

Rocha, & Ruta,2017). The latter studies find empirical support for a complementary relation between

trade and FDI, which might well include the creation or expansion of global value chains via PTAs.5

The analytical framework of our own paper is closest to Anderson etal.(2019), in which they develop

a structural gravity module for bilateral FDI as part of a larger general equilibrium model. However,

there are four important differences between our paper and theirs. First, our paper fully focuses on the

direct impact of PTAs on bilateral FDI using a partial equilibrium model, whereas they want to obtain

the general equilibrium effects of PTAs on welfare.6 Secondly, we allow for the possibility that the

relative FDI friction costs, that is the multilateral resistance terms with respect to FDI, may differ from

2Combining these concepts, Baldwin and Okubo (2014) create a two-dimensional space that accommodates for all FDI

motives and distinguish six main FDI motives: pure horizontal, pure vertical, pure export-platform, local assembly (tariff-

jumping), resource extraction and networked FDI (global value chains).

3Alfaro and Charlton (2009) identify between vertical and horizontal FDI flows by employing the implicit information in

national input–output tables. An alternative database, which has information at the industry level and has been used in some

papers to estimate FDI gravity models is the fDiMarkets database (www.fdima rkets.com) of the Financial Times. However,

this database has two main limitations: it only reports announced (instead of actual) investments and it only has information

on green-field investment rather than on total FDI. These serious database limitations do not allow for the country-wide

analysis we conduct in this paper.

4In particular, our main estimations employ the inward FDI stock data, which fluctuate less and are in general more reliable

than year-to-year FDI flow data. Nonetheless, we also use inward FDI flows to assess the robustness of our results.

5Osnago etal.(2017), in particular, test the effect of deep PTAs on vertical FDI, and not surprisingly, they find a positive

relation. Like our paper, they also focus on the depth of PTAs. However, we use a global database that includes both

horizontal and vertical bilateral FDI.

6The aim of the Anderson etal.(2019, 2020) papers is to obtain the PTAs' general equilibrium effects through trade and FDI

effects on real income. We want to focus on the effect of PTAs on FDI, and we are not concerned with the trade nor the

general equilibrium effects that translate into welfare changes.

|

3205

KOX and ROJaS-ROMaGOSa

the bilateral trade-costs frictions. In Anderson etal.(2019), the multilateral resistance terms are exclu-

sively based on the bilateral trade-cost frictions. Following Bergstrand and Egger (2007) and

Bergstrand and Egger (2010), both trade and FDI flows are driven by a ‘common process’. According

to these authors, this implies that including trade flows as explanatory variables in a (partial equilib-

rium) gravity equation for FDI flows (or vice versa) is a serious mis-specification of the estimations

and it is likely to create an endogeneity bias. Therefore, we do not use trade flows in our FDI gravity

estimations. Employing only FDI data is sufficient for our purposes to assess whether PTAs have a

significant impact on bilateral FDI patterns. Thirdly, we allow for the heterogeneity (depth) of PTAs,

while they only include a dummy variable for the presence of any PTA. The fourth and final difference

is that we generate a larger number of sensitivity analyses to confirm the robustness of the effect of

PTAs on FDI.

The contribution of this paper is that we present a stand-alone partial equilibrium model explaining

bilateral FDI patterns. We empirically estimate the sign and magnitude of the effect that preferential

trade agreements have on FDI, conditional on the depth of the PTA and the presence of other bilat-

eral or multilateral policies—for example being member of the EU single market, the presence of a

bilateral investment treaty. We find that PTAs have a significant and positive effect on bilateral FDI of

around 30%. This effect is slightly larger (35%) for the deepest PTAs, but this result is not statistically

different from just having any PTA—irrespective of its depth—and it is not robust in many alterna-

tive specifications. Therefore, an important finding of this paper is that we do not find that signing a

comprehensive PTA has an empirically larger effect than just signing a ‘shallow’ PTA. On the other

hand, we find that country-pairs that belong to the EU single market experience a very substantial

increase in bilateral FDI of around 135%. Therefore, we argue that it is crucial to control for the EU

single market, when estimating the impact of deep PTAs. Otherwise, the effect of deep PTAs (when

assuming that the EU single market is a deep PTA) is overstated. Finally, we find that the sign and

magnitude of the average PTA effect, of being member of the EU single market, and of participating in

a bilateral investment treaty (BIT), are robust to different econometric specifications. Indirectly, these

results can be interpreted as pointing to a higher relative importance of vertical over horizontal FDI,

by associating the trade cost reductions of PTAs to the nature of the PTA-FDI relation.

This paper is organised as follows. In Section2, we survey the theoretical background linking

PTAs with FDI. Section3 explains our analytical framework and econometric specification. The data

are presented in Section4, and our main structural gravity results are shown in Section5. In Section6,

we run several sensitivity analysis and we conclude in Section7.

2

|

THEORETICAL BACKGROUND

Reviewing the literature on FDI determinants, Blonigen (2005) remarks: ‘Ideally, the FDI literature

would have an established model and empirical specification that lays out the primary long-run deter-

minants of FDI location’. He concludes however that such a model does not yet exist. Given the suc-

cess of the gravity trade model, one wonders why it has not yet become part of the standard empirical

toolkit for the analysis of international patterns in bilateral FDI. In a recent publication, Allen,

Arkolakis, and Takahashi (2020) generalise the gravity trade model by unifying a large set of trade

and geography models, showing that the properties of models within this framework depend crucially

on the value of two gravity constants (aggregate supply and demand elasticities). While their publica-

tion carries the title ‘Universal Gravity’, they do not extend their theory to FDI. This might not be a

coincidence. In this context, Blonigen (2005) again asserts: ‘As with trade flows, a gravity specifica-

tion actually fits cross-country data on FDI reasonably well. However, there is no similar paper to