The impact of trade costs on the European Regional Trade Network: An empirical and theoretical analysis

• Author: Pasquale Commendatore,Luca De Benedictis,Ingrid Kubin,Roberto Basile
• Published date: 01 August 2018
• Date: 01 August 2018
• DOI: http://doi.org/10.1111/roie.12314

SPECIAL ISSUE PAPER

The impact of trade costs on the European Regional

Trade Network: An empirical and theoretical

analysis

Roberto Basile

1

|

Pasquale Commendatore

2

|

Luca De Benedictis

3

|

Ingrid Kubin

4

1

University of Campania “Luigi

Vanvitelli”, Caserta, Italy

2

University of Naples ‘Federico II’,

Naples, Italy

3

University of Macerata, Macerata,Italy

4

WU Vienna University of Economics

and Business, Vienna, Austria

Correspondence

Luca De Benedictis, University of

Macerata, Macerata, Italy.

Email: luca.debenedictis@unimc.it

Abstract

Using intra-European interregional trade data, we analyze

the topology of the E.U. regional trade network. A triad cen-

sus analysis confirms the intuition that the interregional

trade network (and, thus, the European economic integra-

tion) is far from being complete. The majority of the E.U.

interregional trade patterns are characterized by simple, at

best bilateral, configurations. Moreover, we analyze the

effect of trade costs in shaping the topological structure of

the network. It emerges that the relative presence of simple

trade configurations increases with distance, while the rela-

tive presence of more complex trade configurations

decreases with distance. Finally, we discuss the theoretical

underpinnings of these empirical facts through a simple new

economic geography model with three regions. In this

model, we analyze how trade costs shape the pattern of the

trade network. On the whole we find a correspondence

between theoretic and empirical results. However, details

differ and they suggest directions for further research.

1

|

INTRODUCTION

Intra-European trade is usually analyzed at the country level. Lack of intranational and international

regional trade data has prevented, until 2015, a clear description of regional trade flows in Europe. In

this paper, exploiting the availability of new European regional trade data at bilateral level (Thissen,

Lankhuizen, & Jonkeren, 2015), and integrating network analysis and new economic geography mod-

eling, we give evidence of regional trade flows in Europe in 2010, and describe their network structure.

In this paper we extend the analysis started in Basile, Commendatore, De Benedictis, and Kubin

(2016) by focusing on the role played by trade costs in shaping the topological structure of the

network.

578

|

V

C2017 JohnWiley & Sons Ltd wileyonlinelibrary.com/journal/roie Rev IntEcon. 2018;2 6:578–609.

DOI: 10.1111/roie.12314

Trade costs are notoriously difficult to measure (Anderson & Van Wincoop, 2004), even more so

if the quantification is undertaken at the regional level. The strategy we adopted in our analysis is to

associate the theoretical bilateral trade cost between region rand region s,T

rs

, to a latent variable that

is empirically observed in its realization, through two components: one is the geographical distance

between rand s, and the other is the presence of a border effect that generates a distinction between

intranational and international regional trade.

The empirical investigation of the interregional trade network in Europe focuses on the results of a

triad census analysis. First, it emerges that the interregional trade network(and, thus, the European eco-

nomic integration) is far from being complete. The majority of the E.U. interregional trade patterns are

characterized by simple, at best bilateral, configurations; and, when two regions establish a mutual

trade relationship, this fosters them to export to, more than to import from, a third region. Then, by

decomposing the overall interregional trade network graph in different subgraphs according to the

quantile distribution of bilateral distances, we also analyze the effect of trade costs in shaping the topo-

logical structure of the network. From this analysis it emerges that the relative presence of simple trade

configurations increases with distance, while the relative presence of more complex trade configura-

tions decreases with distance. Finally, we consider a further dimension of trade costs by applying triad

census analysis to regions within the same country (domestic trade) and regions of different countries

(international trade). The triad census analysis applied to the corresponding subgraphs provides evi-

dence of the occurrence of a border effect: bilateral trade flows are relatively more frequent between

regions of the same country rather than between regions of different countries. In contrast, hierarchical

structures, such as mutual stars, are relatively more frequent when only regions of different countries

are selected, confirming the important role of hubs or export platforms played by core regions such as

^

Ile-de-France, Inner London, and Madrid.

Then, we discuss the theoretical underpinnings of these empirical facts through a three-region new

economic geography (NEG) model that extends previous analyses (see Ottaviano, Tabuchi, & Thisse,

2002; Behrens, 2011; Ago, Isono, & Tabuchi, 2006; Basile et al., 2016). In Basile et al. (2016) we put

forward a three-region NEG model that is more general than those previously existing in the literature,

where the distance between the regions is not necessarily the same. Allowing for the possible presence

or absence of trade links between the regions gives rise to 64 possible network structures. Without fur-

ther specifications, these structures can be grouped into 16 isomorphic classes. In that contribution, we

mostly limited our analysis to the conditions, expressed in terms of different combinations of trade

costs and spatial factor allocation, corresponding to each of those classes. In the present article, we

explicitly number the regions and specify their distances losing the isomorphisms between the network

structures. We derive analytic conditions for the presence of trade links between regions and study

how changes in trade costs affect the occurrence of the 64 possible trade network structures. Based on

Proposition 1, we argue that some of these configurations (corresponding to the 16 different types of

the triad census analysis) can be discarded a priori. Moreover, for each type of network structure or

triad we establish its likelihood, given trade distances and the distribution of the economic activity (via

conjecture 1 and simulations analysis). It turns out that more connected structures are a direct conse-

quence of lower trade costs. We verify that, owing to competition, close regions engage more easily in

mutual trade, and are less open to trade with more distant regions.

Finally, we use the theoretical analysis to shed some light on the empirical results looking at the

impact of trade costs on the frequency of trade network structures. On the whole, we find a correspon-

dence between theoretic and empirical results. However, details differ and they suggest directions for

further research.

The paper is structured as follows: Section 2 presents the results of the empirical analysis, Section

3 develops the theoretical model, and Section 4 provides a discussion of the results and conclusions.

BASILE ET AL.

|

579

2

|

EMPIRICAL ANALYSIS

2.1

|

E.U. regional trade data

The data produced by the PBL Netherlands Environmental Assessment Agency (Thissen et al., 2015)

is one of the first attempts to give account of E.U. regional trade. The data, obtained by breaking down

international trade flows and national supply and use tables to the regional level,

1

is used by the E.U.

Commission for regional policy simulations. Unlike gravity model estimations, the methodology used

to calculate regional trade flows at the bilateral level stays as close as possible to observed data, with-

out imposing any trade pattern constraint. Nevertheless, as pointed out by Thissen, Diodato, & Van

Oort (2013), one has to keep in mind that the constructed interregional trade data are inferred from

other data sources and are not measured as a flow from one region to another. Given the compatibility

constraints with macro variables, some bias in the trade flows between regions inside a country or

between regions of different countries might result from the weighting procedure used in the construc-

tion of the data. In order to reduce this bias, the original Project Based Learning (PBL) database has

been adjusted to make the trade flows consistent with country totals from the national social accounting

matrices. This is done by a generalized RAS method, which keeps the structure of bilateral trade flows

as much unchanged as possible. The resulting data can therefore be used as such to describe and obtain

information on the topology of the E.U. regional trade network. Moreover, as it will be clarified in

Subsection 2.2, we only exploit the information on the values of the flows to identify a proper thresh-

old value, according to which we transform the import–export matrix in a binary network matrix; and

the last one will be the basic information used in the triad census analysis.

The original data used in our empirical analysis contains information, only for the year 2010, on

bilateral trade flows (priced free on board and measured in millions of euros) between 268 regions

belonging to 28 European countries.

2

Export and import flows refer to an aggregate category of sec-

tors, which includes “Manufacturing”and two more sectors: “Mining and quarrying,”and “Electricity,

gas and water supply.”From the original data we excluded the domestic use of domestic production,

also known as the regional trade flow from region sto region s, imputing a zero-trade flow on the

main diagonal of the trade matrix (268 3268571,824 dyadic observations).

A descriptive analysis of the trade network and its visualization are reported in Appendix A; further

information is also reported in Basile et al. (2016) for a slightly different version of the data.

3

In what fol-

lows, we focus on the topological properties of the network. Specifically, we start from a triad census

analysis applied to the overall interregional trade network (Subsection 2.2); then we explore the role of

physical distance and border effects in shaping the topological structure of the network (Subsection 2.3).

2.2

|

Triad census

The structure of the European regional trade network can be studied through many different network

analysis tools (Newman, 2003; Jackson, 2008; Hanneman & Riddle, 2005). We are interested, in par-

ticular, in the topological properties of the network concerning the relation between three nodes. Three

is the minimum number to study interdependence. If the decision to trade from rto sis taken consider-

ing opportunity costs, the option of trading with an alternative region must be considered.

More specifically, we follow Basile et al. (2016) in studying the structure of the regional trade net-

work focusing on the triad as the unit of analysis. The “Triad Census”(the count of the various types

of triads in the network) is a classical network analysis tool (Wasserman and Faust, 1994), usually

applied to a single, directed and binary network, and we follow the tradition in this respect. Taking the

three nodes V

s

,V

r

,andV

k

,wheres6¼ r6¼ k, we can call them a “triple.”If we also consider the pres-

ence or absence of links between these different nodes, we have a triad (T

srk

). If the network is

580

|

BASILE ET AL.